Merge pull request #259 from yhcao6/dcn_cpp_extension

Deformable Convolution v2 cpp extension

MODEL_ZOO.md

@@ -184,6 +184,28 @@ We released RPN, Faster R-CNN and Mask R-CNN models in the first version. More m
 - The `3x` schedule is epoch [28, 34, 36].
 - The memory is measured with `torch.cuda.max_memory_allocated()` instead of `torch.cuda.max_memory_cached()`. We will update the memory usage of other models in the future.
 
+### Deformable Convolution v2
+
+| Backbone  | Model        | Style   | Conv          | Pool   | Lr schd | Mem (GB) | Train time (s/iter) | Inf time (fps) | box AP | mask AP | Download |
+|:---------:|:------------:|:-------:|:-------------:|:------:|:-------:|:--------:|:-------------------:|:--------------:|:------:|:-------:|:--------:|
+| R-50-FPN  | Faster       | pytorch | dconv(c3-c5)  | -      | 1x      | 3.9      | 0.594               | 10.2           | 40.0   | -       | [model](https://s3.ap-northeast-2.amazonaws.com/open-mmlab/mmdetection/models/dcn/faster_rcnn_dconv_c3-c5_r50_fpn_1x_20190125-e41688c9.pth) |
+| R-50-FPN  | Faster       | pytorch | mdconv(c3-c5) | -      | 1x      | 3.7      | 0.598               | 10.0           | 40.3   | -       | [model](https://s3.ap-northeast-2.amazonaws.com/open-mmlab/mmdetection/models/dcn/faster_rcnn_mdconv_c3-c5_r50_fpn_1x_20190125-1b768045.pth) |
+| R-50-FPN  | Faster       | pytorch | -             | dpool  | 1x      | 4.6      | 0.714               | 8.7            | 37.9   | -       | [model](https://s3.ap-northeast-2.amazonaws.com/open-mmlab/mmdetection/models/dcn/faster_rcnn_dpool_r50_fpn_1x_20190125-f4fc1d70.pth) |
+| R-50-FPN  | Faster       | pytorch | -             | mdpool | 1x      | 5.2      | 0.769               | 8.2            | 38.1   | -       | [model](https://s3.ap-northeast-2.amazonaws.com/open-mmlab/mmdetection/models/dcn/faster_rcnn_mdpool_r50_fpn_1x_20190125-473d0f3d.pth) |
+| R-101-FPN | Faster       | pytorch | dconv(c3-c5)  | -      | 1x      | 5.8      | 0.811               | 8.0            | 42.1   | -       | [model](https://s3.ap-northeast-2.amazonaws.com/open-mmlab/mmdetection/models/dcn/faster_rcnn_dconv_c3-c5_r101_fpn_1x_20190125-a7e31b65.pth) |
+| R-50-FPN  | Mask         | pytorch | dconv(c3-c5)  | -      | 1x      | 4.5      | 0.712               | 7.7            | 41.1   | 37.2    | [model](https://s3.ap-northeast-2.amazonaws.com/open-mmlab/mmdetection/models/dcn/mask_rcnn_dconv_c3-c5_r50_fpn_1x_20190125-4f94ff79.pth) |
+| R-50-FPN  | Mask         | pytorch | mdconv(c3-c5) | -      | 1x      | 4.5      | 0.712               | 7.7            | 41.4   | 37.4    | [model](https://s3.ap-northeast-2.amazonaws.com/open-mmlab/mmdetection/models/dcn/mask_rcnn_mdconv_c3-c5_r50_fpn_1x_20190125-c5601dc3.pth) |
+| R-101-FPN | Mask         | pytorch | dconv(c3-c5)  | -      | 1x      | 6.4      | 0.939               | 6.5            | 43.2   | 38.7    | [model](https://s3.ap-northeast-2.amazonaws.com/open-mmlab/mmdetection/models/dcn/mask_rcnn_dconv_c3-c5_r101_fpn_1x_20190125-decb6db5.pth) |
+| R-50-FPN  | Cascade      | pytorch | dconv(c3-c5)  | -      | 1x      | 4.4      | 0.660               | 7.6            | 44.1   | -       | [model](https://s3.ap-northeast-2.amazonaws.com/open-mmlab/mmdetection/models/dcn/cascade_rcnn_dconv_c3-c5_r50_fpn_1x_20190125-dfa53166.pth) |
+| R-101-FPN | Cascade      | pytorch | dconv(c3-c5)  | -      | 1x      | 6.3      | 0.881               | 6.8            | 45.1   | -       | [model](https://s3.ap-northeast-2.amazonaws.com/open-mmlab/mmdetection/models/dcn/cascade_rcnn_dconv_c3-c5_r101_fpn_1x_20190125-aaa877cc.pth) |
+| R-50-FPN  | Cascade Mask | pytorch | dconv(c3-c5)  | -      | 1x      | 6.6      | 0.942               | 5.7            | 44.5   | 38.3    | [model](https://s3.ap-northeast-2.amazonaws.com/open-mmlab/mmdetection/models/dcn/cascade_mask_rcnn_dconv_c3-c5_r50_fpn_1x_20190125-09d8a443.pth) |
+| R-101-FPN | Cascade Mask | pytorch | dconv(c3-c5)  | -      | 1x      | 8.5      | 1.156               | 5.1            | 45.8   | 39.5    | [model](https://s3.ap-northeast-2.amazonaws.com/open-mmlab/mmdetection/models/dcn/cascade_mask_rcnn_dconv_c3-c5_r101_fpn_1x_20190125-0d62c190.pth) |
+
+**Notes:**
+
+- `dconv` and `mdconv` denote (modulated) deformable convolution, `c3-c5` means adding dconv in resnet stage 3 to 5. `dpool` and `mdpool` denote (modulated) deformable roi pooling.
+- The memory is measured with `torch.cuda.max_memory_allocated()`. The batch size is 16 (2 images per GPU).
+- The dcn ops are modified from https://github.com/chengdazhi/Deformable-Convolution-V2-PyTorch, which should be more memory efficient and slightly faster.
 
 ## Comparison with Detectron
 

README.md

@@ -78,6 +78,7 @@ Results and models are available in the [Model zoo](MODEL_ZOO.md).
 | RetinaNet          | ✓        | ✓        | ☐        | ✗        |
 
 Other features
+- [x] DCNv2
 - [x] Group Normalization
 - [x] OHEM
 - [x] Soft-NMS

compile.sh

@@ -20,3 +20,10 @@ echo "Building nms op..."
 cd ../nms
 make clean
 make PYTHON=${PYTHON}
+
+echo "Building dcn..."
+cd ../dcn
+if [ -d "build" ]; then
+    rm -r build
+fi
+$PYTHON setup.py build_ext --inplace

configs/dcn/cascade_mask_rcnn_dconv_c3-c5_r50_fpn_1x.py

+# model settings
+model = dict(
+    type='CascadeRCNN',
+    num_stages=3,
+    pretrained='modelzoo://resnet50',
+    backbone=dict(
+        type='ResNet',
+        depth=50,
+        num_stages=4,
+        out_indices=(0, 1, 2, 3),
+        frozen_stages=1,
+        style='pytorch',
+        dcn=dict(
+            modulated=False,
+            deformable_groups=1,
+            fallback_on_stride=False),
+        stage_with_dcn=(False, True, True, True)),
+    neck=dict(
+        type='FPN',
+        in_channels=[256, 512, 1024, 2048],
+        out_channels=256,
+        num_outs=5),
+    rpn_head=dict(
+        type='RPNHead',
+        in_channels=256,
+        feat_channels=256,
+        anchor_scales=[8],
+        anchor_ratios=[0.5, 1.0, 2.0],
+        anchor_strides=[4, 8, 16, 32, 64],
+        target_means=[.0, .0, .0, .0],
+        target_stds=[1.0, 1.0, 1.0, 1.0],
+        use_sigmoid_cls=True),
+    bbox_roi_extractor=dict(
+        type='SingleRoIExtractor',
+        roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
+        out_channels=256,
+        featmap_strides=[4, 8, 16, 32]),
+    bbox_head=[
+        dict(
+            type='SharedFCBBoxHead',
+            num_fcs=2,
+            in_channels=256,
+            fc_out_channels=1024,
+            roi_feat_size=7,
+            num_classes=81,
+            target_means=[0., 0., 0., 0.],
+            target_stds=[0.1, 0.1, 0.2, 0.2],
+            reg_class_agnostic=True),
+        dict(
+            type='SharedFCBBoxHead',
+            num_fcs=2,
+            in_channels=256,
+            fc_out_channels=1024,
+            roi_feat_size=7,
+            num_classes=81,
+            target_means=[0., 0., 0., 0.],
+            target_stds=[0.05, 0.05, 0.1, 0.1],
+            reg_class_agnostic=True),
+        dict(
+            type='SharedFCBBoxHead',
+            num_fcs=2,
+            in_channels=256,
+            fc_out_channels=1024,
+            roi_feat_size=7,
+            num_classes=81,
+            target_means=[0., 0., 0., 0.],
+            target_stds=[0.033, 0.033, 0.067, 0.067],
+            reg_class_agnostic=True)
+    ],
+    mask_roi_extractor=dict(
+        type='SingleRoIExtractor',
+        roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
+        out_channels=256,
+        featmap_strides=[4, 8, 16, 32]),
+    mask_head=dict(
+        type='FCNMaskHead',
+        num_convs=4,
+        in_channels=256,
+        conv_out_channels=256,
+        num_classes=81))
+# model training and testing settings
+train_cfg = dict(
+    rpn=dict(
+        assigner=dict(
+            type='MaxIoUAssigner',
+            pos_iou_thr=0.7,
+            neg_iou_thr=0.3,
+            min_pos_iou=0.3,
+            ignore_iof_thr=-1),
+        sampler=dict(
+            type='RandomSampler',
+            num=256,
+            pos_fraction=0.5,
+            neg_pos_ub=-1,
+            add_gt_as_proposals=False),
+        allowed_border=0,
+        pos_weight=-1,
+        smoothl1_beta=1 / 9.0,
+        debug=False),
+    rcnn=[
+        dict(
+            assigner=dict(
+                type='MaxIoUAssigner',
+                pos_iou_thr=0.5,
+                neg_iou_thr=0.5,
+                min_pos_iou=0.5,
+                ignore_iof_thr=-1),
+            sampler=dict(
+                type='RandomSampler',
+                num=512,
+                pos_fraction=0.25,
+                neg_pos_ub=-1,
+                add_gt_as_proposals=True),
+            mask_size=28,
+            pos_weight=-1,
+            debug=False),
+        dict(
+            assigner=dict(
+                type='MaxIoUAssigner',
+                pos_iou_thr=0.6,
+                neg_iou_thr=0.6,
+                min_pos_iou=0.6,
+                ignore_iof_thr=-1),
+            sampler=dict(
+                type='RandomSampler',
+                num=512,
+                pos_fraction=0.25,
+                neg_pos_ub=-1,
+                add_gt_as_proposals=True),
+            mask_size=28,
+            pos_weight=-1,
+            debug=False),
+        dict(
+            assigner=dict(
+                type='MaxIoUAssigner',
+                pos_iou_thr=0.7,
+                neg_iou_thr=0.7,
+                min_pos_iou=0.7,
+                ignore_iof_thr=-1),
+            sampler=dict(
+                type='RandomSampler',
+                num=512,
+                pos_fraction=0.25,
+                neg_pos_ub=-1,
+                add_gt_as_proposals=True),
+            mask_size=28,
+            pos_weight=-1,
+            debug=False)
+    ],
+    stage_loss_weights=[1, 0.5, 0.25])
+test_cfg = dict(
+    rpn=dict(
+        nms_across_levels=False,
+        nms_pre=2000,
+        nms_post=2000,
+        max_num=2000,
+        nms_thr=0.7,
+        min_bbox_size=0),
+    rcnn=dict(
+        score_thr=0.05,
+        nms=dict(type='nms', iou_thr=0.5),
+        max_per_img=100,
+        mask_thr_binary=0.5),
+    keep_all_stages=False)
+# dataset settings
+dataset_type = 'CocoDataset'
+data_root = 'data/coco/'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+data = dict(
+    imgs_per_gpu=2,
+    workers_per_gpu=2,
+    train=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_train2017.json',
+        img_prefix=data_root + 'train2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0.5,
+        with_mask=True,
+        with_crowd=True,
+        with_label=True),
+    val=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0,
+        with_mask=True,
+        with_crowd=True,
+        with_label=True),
+    test=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0,
+        with_mask=True,
+        with_label=False,
+        test_mode=True))
+# optimizer
+optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=1.0 / 3,
+    step=[8, 11])
+checkpoint_config = dict(interval=1)
+# yapf:disable
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        # dict(type='TensorboardLoggerHook')
+    ])
+# yapf:enable
+# runtime settings
+total_epochs = 12
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+work_dir = './work_dirs/cascade_mask_rcnn_dconv_c3-c5_r50_fpn_1x'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]

configs/dcn/cascade_rcnn_dconv_c3-c5_r50_fpn_1x.py

+# model settings
+model = dict(
+    type='CascadeRCNN',
+    num_stages=3,
+    pretrained='modelzoo://resnet50',
+    backbone=dict(
+        type='ResNet',
+        depth=50,
+        num_stages=4,
+        out_indices=(0, 1, 2, 3),
+        frozen_stages=1,
+        style='pytorch',
+        dcn=dict(
+            modulated=False,
+            deformable_groups=1,
+            fallback_on_stride=False),
+        stage_with_dcn=(False, True, True, True)),
+    neck=dict(
+        type='FPN',
+        in_channels=[256, 512, 1024, 2048],
+        out_channels=256,
+        num_outs=5),
+    rpn_head=dict(
+        type='RPNHead',
+        in_channels=256,
+        feat_channels=256,
+        anchor_scales=[8],
+        anchor_ratios=[0.5, 1.0, 2.0],
+        anchor_strides=[4, 8, 16, 32, 64],
+        target_means=[.0, .0, .0, .0],
+        target_stds=[1.0, 1.0, 1.0, 1.0],
+        use_sigmoid_cls=True),
+    bbox_roi_extractor=dict(
+        type='SingleRoIExtractor',
+        roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
+        out_channels=256,
+        featmap_strides=[4, 8, 16, 32]),
+    bbox_head=[
+        dict(
+            type='SharedFCBBoxHead',
+            num_fcs=2,
+            in_channels=256,
+            fc_out_channels=1024,
+            roi_feat_size=7,
+            num_classes=81,
+            target_means=[0., 0., 0., 0.],
+            target_stds=[0.1, 0.1, 0.2, 0.2],
+            reg_class_agnostic=True),
+        dict(
+            type='SharedFCBBoxHead',
+            num_fcs=2,
+            in_channels=256,
+            fc_out_channels=1024,
+            roi_feat_size=7,
+            num_classes=81,
+            target_means=[0., 0., 0., 0.],
+            target_stds=[0.05, 0.05, 0.1, 0.1],
+            reg_class_agnostic=True),
+        dict(
+            type='SharedFCBBoxHead',
+            num_fcs=2,
+            in_channels=256,
+            fc_out_channels=1024,
+            roi_feat_size=7,
+            num_classes=81,
+            target_means=[0., 0., 0., 0.],
+            target_stds=[0.033, 0.033, 0.067, 0.067],
+            reg_class_agnostic=True)
+    ])
+# model training and testing settings
+train_cfg = dict(
+    rpn=dict(
+        assigner=dict(
+            type='MaxIoUAssigner',
+            pos_iou_thr=0.7,
+            neg_iou_thr=0.3,
+            min_pos_iou=0.3,
+            ignore_iof_thr=-1),
+        sampler=dict(
+            type='RandomSampler',
+            num=256,
+            pos_fraction=0.5,
+            neg_pos_ub=-1,
+            add_gt_as_proposals=False),
+        allowed_border=0,
+        pos_weight=-1,
+        smoothl1_beta=1 / 9.0,
+        debug=False),
+    rcnn=[
+        dict(
+            assigner=dict(
+                type='MaxIoUAssigner',
+                pos_iou_thr=0.5,
+                neg_iou_thr=0.5,
+                min_pos_iou=0.5,
+                ignore_iof_thr=-1),
+            sampler=dict(
+                type='RandomSampler',
+                num=512,
+                pos_fraction=0.25,
+                neg_pos_ub=-1,
+                add_gt_as_proposals=True),
+            pos_weight=-1,
+            debug=False),
+        dict(
+            assigner=dict(
+                type='MaxIoUAssigner',
+                pos_iou_thr=0.6,
+                neg_iou_thr=0.6,
+                min_pos_iou=0.6,
+                ignore_iof_thr=-1),
+            sampler=dict(
+                type='RandomSampler',
+                num=512,
+                pos_fraction=0.25,
+                neg_pos_ub=-1,
+                add_gt_as_proposals=True),
+            pos_weight=-1,
+            debug=False),
+        dict(
+            assigner=dict(
+                type='MaxIoUAssigner',
+                pos_iou_thr=0.7,
+                neg_iou_thr=0.7,
+                min_pos_iou=0.7,
+                ignore_iof_thr=-1),
+            sampler=dict(
+                type='RandomSampler',
+                num=512,
+                pos_fraction=0.25,
+                neg_pos_ub=-1,
+                add_gt_as_proposals=True),
+            pos_weight=-1,
+            debug=False)
+    ],
+    stage_loss_weights=[1, 0.5, 0.25])
+test_cfg = dict(
+    rpn=dict(
+        nms_across_levels=False,
+        nms_pre=2000,
+        nms_post=2000,
+        max_num=2000,
+        nms_thr=0.7,
+        min_bbox_size=0),
+    rcnn=dict(
+        score_thr=0.05, nms=dict(type='nms', iou_thr=0.5), max_per_img=100),
+    keep_all_stages=False)
+# dataset settings
+dataset_type = 'CocoDataset'
+data_root = 'data/coco/'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+data = dict(
+    imgs_per_gpu=2,
+    workers_per_gpu=2,
+    train=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_train2017.json',
+        img_prefix=data_root + 'train2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0.5,
+        with_mask=False,
+        with_crowd=True,
+        with_label=True),
+    val=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0,
+        with_mask=False,
+        with_crowd=True,
+        with_label=True),
+    test=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0,
+        with_mask=False,
+        with_label=False,
+        test_mode=True))
+# optimizer
+optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=1.0 / 3,
+    step=[8, 11])
+checkpoint_config = dict(interval=1)
+# yapf:disable
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        # dict(type='TensorboardLoggerHook')
+    ])
+# yapf:enable
+# runtime settings
+total_epochs = 12
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+work_dir = './work_dirs/cascade_rcnn_dconv_c3-c5_r50_fpn_1x'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]

configs/dcn/faster_rcnn_dconv_c3-c5_r50_fpn_1x.py

+# model settings
+model = dict(
+    type='FasterRCNN',
+    pretrained='modelzoo://resnet50',
+    backbone=dict(
+        type='ResNet',
+        depth=50,
+        num_stages=4,
+        out_indices=(0, 1, 2, 3),
+        frozen_stages=1,
+        style='pytorch',
+        dcn=dict(
+            modulated=False,
+            deformable_groups=1,
+            fallback_on_stride=False),
+        stage_with_dcn=(False, True, True, True)),
+    neck=dict(
+        type='FPN',
+        in_channels=[256, 512, 1024, 2048],
+        out_channels=256,
+        num_outs=5),
+    rpn_head=dict(
+        type='RPNHead',
+        in_channels=256,
+        feat_channels=256,
+        anchor_scales=[8],
+        anchor_ratios=[0.5, 1.0, 2.0],
+        anchor_strides=[4, 8, 16, 32, 64],
+        target_means=[.0, .0, .0, .0],
+        target_stds=[1.0, 1.0, 1.0, 1.0],
+        use_sigmoid_cls=True),
+    bbox_roi_extractor=dict(
+        type='SingleRoIExtractor',
+        roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
+        out_channels=256,
+        featmap_strides=[4, 8, 16, 32]),
+    bbox_head=dict(
+        type='SharedFCBBoxHead',
+        num_fcs=2,
+        in_channels=256,
+        fc_out_channels=1024,
+        roi_feat_size=7,
+        num_classes=81,
+        target_means=[0., 0., 0., 0.],
+        target_stds=[0.1, 0.1, 0.2, 0.2],
+        reg_class_agnostic=False))
+# model training and testing settings
+train_cfg = dict(
+    rpn=dict(
+        assigner=dict(
+            type='MaxIoUAssigner',
+            pos_iou_thr=0.7,
+            neg_iou_thr=0.3,
+            min_pos_iou=0.3,
+            ignore_iof_thr=-1),
+        sampler=dict(
+            type='RandomSampler',
+            num=256,
+            pos_fraction=0.5,
+            neg_pos_ub=-1,
+            add_gt_as_proposals=False),
+        allowed_border=0,
+        pos_weight=-1,
+        smoothl1_beta=1 / 9.0,
+        debug=False),
+    rcnn=dict(
+        assigner=dict(
+            type='MaxIoUAssigner',
+            pos_iou_thr=0.5,
+            neg_iou_thr=0.5,
+            min_pos_iou=0.5,
+            ignore_iof_thr=-1),
+        sampler=dict(
+            type='RandomSampler',
+            num=512,
+            pos_fraction=0.25,
+            neg_pos_ub=-1,
+            add_gt_as_proposals=True),
+        pos_weight=-1,
+        debug=False))
+test_cfg = dict(
+    rpn=dict(
+        nms_across_levels=False,
+        nms_pre=2000,
+        nms_post=2000,
+        max_num=2000,
+        nms_thr=0.7,
+        min_bbox_size=0),
+    rcnn=dict(
+        score_thr=0.05, nms=dict(type='nms', iou_thr=0.5), max_per_img=100)
+    # soft-nms is also supported for rcnn testing
+    # e.g., nms=dict(type='soft_nms', iou_thr=0.5, min_score=0.05)
+)
+# dataset settings
+dataset_type = 'CocoDataset'
+data_root = 'data/coco/'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+data = dict(
+    imgs_per_gpu=2,
+    workers_per_gpu=2,
+    train=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_train2017.json',
+        img_prefix=data_root + 'train2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0.5,
+        with_mask=False,
+        with_crowd=True,
+        with_label=True),
+    val=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0,
+        with_mask=False,
+        with_crowd=True,
+        with_label=True),
+    test=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0,
+        with_mask=False,
+        with_label=False,
+        test_mode=True))
+# optimizer
+optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=1.0 / 3,
+    step=[8, 11])
+checkpoint_config = dict(interval=1)
+# yapf:disable
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        # dict(type='TensorboardLoggerHook')
+    ])
+# yapf:enable
+# runtime settings
+total_epochs = 12
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+work_dir = './work_dirs/faster_rcnn_dconv_c3-c5_r50_fpn_1x'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]

configs/dcn/faster_rcnn_dpool_r50_fpn_1x.py

+# model settings
+model = dict(
+    type='FasterRCNN',
+    pretrained='modelzoo://resnet50',
+    backbone=dict(
+        type='ResNet',
+        depth=50,
+        num_stages=4,
+        out_indices=(0, 1, 2, 3),
+        frozen_stages=1,
+        style='pytorch'),
+    neck=dict(
+        type='FPN',
+        in_channels=[256, 512, 1024, 2048],
+        out_channels=256,
+        num_outs=5),
+    rpn_head=dict(
+        type='RPNHead',
+        in_channels=256,
+        feat_channels=256,
+        anchor_scales=[8],
+        anchor_ratios=[0.5, 1.0, 2.0],
+        anchor_strides=[4, 8, 16, 32, 64],
+        target_means=[.0, .0, .0, .0],
+        target_stds=[1.0, 1.0, 1.0, 1.0],
+        use_sigmoid_cls=True),
+    bbox_roi_extractor=dict(
+        type='SingleRoIExtractor',
+        roi_layer=dict(
+            type='DeformRoIPoolingPack',
+            out_size=7,
+            out_channels=256,
+            no_trans=False,
+            group_size=1,
+            trans_std=0.1),
+        out_channels=256,
+        featmap_strides=[4, 8, 16, 32]),
+    bbox_head=dict(
+        type='SharedFCBBoxHead',
+        num_fcs=2,
+        in_channels=256,
+        fc_out_channels=1024,
+        roi_feat_size=7,
+        num_classes=81,
+        target_means=[0., 0., 0., 0.],
+        target_stds=[0.1, 0.1, 0.2, 0.2],
+        reg_class_agnostic=False))
+# model training and testing settings
+train_cfg = dict(
+    rpn=dict(
+        assigner=dict(
+            type='MaxIoUAssigner',
+            pos_iou_thr=0.7,
+            neg_iou_thr=0.3,
+            min_pos_iou=0.3,
+            ignore_iof_thr=-1),
+        sampler=dict(
+            type='RandomSampler',
+            num=256,
+            pos_fraction=0.5,
+            neg_pos_ub=-1,
+            add_gt_as_proposals=False),
+        allowed_border=0,
+        pos_weight=-1,
+        smoothl1_beta=1 / 9.0,
+        debug=False),
+    rcnn=dict(
+        assigner=dict(
+            type='MaxIoUAssigner',
+            pos_iou_thr=0.5,
+            neg_iou_thr=0.5,
+            min_pos_iou=0.5,
+            ignore_iof_thr=-1),
+        sampler=dict(
+            type='RandomSampler',
+            num=512,
+            pos_fraction=0.25,
+            neg_pos_ub=-1,
+            add_gt_as_proposals=True),
+        pos_weight=-1,
+        debug=False))
+test_cfg = dict(
+    rpn=dict(
+        nms_across_levels=False,
+        nms_pre=2000,
+        nms_post=2000,
+        max_num=2000,
+        nms_thr=0.7,
+        min_bbox_size=0),
+    rcnn=dict(
+        score_thr=0.05, nms=dict(type='nms', iou_thr=0.5), max_per_img=100)
+    # soft-nms is also supported for rcnn testing
+    # e.g., nms=dict(type='soft_nms', iou_thr=0.5, min_score=0.05)
+)
+# dataset settings
+dataset_type = 'CocoDataset'
+data_root = 'data/coco/'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+data = dict(
+    imgs_per_gpu=2,
+    workers_per_gpu=2,
+    train=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_train2017.json',
+        img_prefix=data_root + 'train2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0.5,
+        with_mask=False,
+        with_crowd=True,
+        with_label=True),
+    val=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0,
+        with_mask=False,
+        with_crowd=True,
+        with_label=True),
+    test=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0,
+        with_mask=False,
+        with_label=False,
+        test_mode=True))
+# optimizer
+optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=1.0 / 3,
+    step=[8, 11])
+checkpoint_config = dict(interval=1)
+# yapf:disable
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        # dict(type='TensorboardLoggerHook')
+    ])
+# yapf:enable
+# runtime settings
+total_epochs = 12
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+work_dir = './work_dirs/faster_rcnn_dpool_r50_fpn_1x'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]

configs/dcn/faster_rcnn_mdconv_c3-c5_r50_fpn_1x.py

+# model settings
+model = dict(
+    type='FasterRCNN',
+    pretrained='modelzoo://resnet50',
+    backbone=dict(
+        type='ResNet',
+        depth=50,
+        num_stages=4,
+        out_indices=(0, 1, 2, 3),
+        frozen_stages=1,
+        style='pytorch',
+        dcn=dict(
+            modulated=True,
+            deformable_groups=1,
+            fallback_on_stride=False),
+        stage_with_dcn=(False, True, True, True)),
+    neck=dict(
+        type='FPN',
+        in_channels=[256, 512, 1024, 2048],
+        out_channels=256,
+        num_outs=5),
+    rpn_head=dict(
+        type='RPNHead',
+        in_channels=256,
+        feat_channels=256,
+        anchor_scales=[8],
+        anchor_ratios=[0.5, 1.0, 2.0],
+        anchor_strides=[4, 8, 16, 32, 64],
+        target_means=[.0, .0, .0, .0],
+        target_stds=[1.0, 1.0, 1.0, 1.0],
+        use_sigmoid_cls=True),
+    bbox_roi_extractor=dict(
+        type='SingleRoIExtractor',
+        roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
+        out_channels=256,
+        featmap_strides=[4, 8, 16, 32]),
+    bbox_head=dict(
+        type='SharedFCBBoxHead',
+        num_fcs=2,
+        in_channels=256,
+        fc_out_channels=1024,
+        roi_feat_size=7,
+        num_classes=81,
+        target_means=[0., 0., 0., 0.],
+        target_stds=[0.1, 0.1, 0.2, 0.2],
+        reg_class_agnostic=False))
+# model training and testing settings
+train_cfg = dict(
+    rpn=dict(
+        assigner=dict(
+            type='MaxIoUAssigner',
+            pos_iou_thr=0.7,
+            neg_iou_thr=0.3,
+            min_pos_iou=0.3,
+            ignore_iof_thr=-1),
+        sampler=dict(
+            type='RandomSampler',
+            num=256,
+            pos_fraction=0.5,
+            neg_pos_ub=-1,
+            add_gt_as_proposals=False),
+        allowed_border=0,
+        pos_weight=-1,
+        smoothl1_beta=1 / 9.0,
+        debug=False),
+    rcnn=dict(
+        assigner=dict(
+            type='MaxIoUAssigner',
+            pos_iou_thr=0.5,
+            neg_iou_thr=0.5,
+            min_pos_iou=0.5,
+            ignore_iof_thr=-1),
+        sampler=dict(
+            type='RandomSampler',
+            num=512,
+            pos_fraction=0.25,
+            neg_pos_ub=-1,
+            add_gt_as_proposals=True),
+        pos_weight=-1,
+        debug=False))
+test_cfg = dict(
+    rpn=dict(
+        nms_across_levels=False,
+        nms_pre=2000,
+        nms_post=2000,
+        max_num=2000,
+        nms_thr=0.7,
+        min_bbox_size=0),
+    rcnn=dict(
+        score_thr=0.05, nms=dict(type='nms', iou_thr=0.5), max_per_img=100)
+    # soft-nms is also supported for rcnn testing
+    # e.g., nms=dict(type='soft_nms', iou_thr=0.5, min_score=0.05)
+)
+# dataset settings
+dataset_type = 'CocoDataset'
+data_root = 'data/coco/'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+data = dict(
+    imgs_per_gpu=2,
+    workers_per_gpu=2,
+    train=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_train2017.json',
+        img_prefix=data_root + 'train2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0.5,
+        with_mask=False,
+        with_crowd=True,
+        with_label=True),
+    val=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0,
+        with_mask=False,
+        with_crowd=True,
+        with_label=True),
+    test=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0,
+        with_mask=False,
+        with_label=False,
+        test_mode=True))
+# optimizer
+optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=1.0 / 3,
+    step=[8, 11])
+checkpoint_config = dict(interval=1)
+# yapf:disable
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        # dict(type='TensorboardLoggerHook')
+    ])
+# yapf:enable
+# runtime settings
+total_epochs = 12
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+work_dir = './work_dirs/faster_rcnn_mdconv_c3-c5_r50_fpn_1x'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]

configs/dcn/faster_rcnn_mdpool_r50_fpn_1x.py

+# model settings
+model = dict(
+    type='FasterRCNN',
+    pretrained='modelzoo://resnet50',
+    backbone=dict(
+        type='ResNet',
+        depth=50,
+        num_stages=4,
+        out_indices=(0, 1, 2, 3),
+        frozen_stages=1,
+        style='pytorch'),
+    neck=dict(
+        type='FPN',
+        in_channels=[256, 512, 1024, 2048],
+        out_channels=256,
+        num_outs=5),
+    rpn_head=dict(
+        type='RPNHead',
+        in_channels=256,
+        feat_channels=256,
+        anchor_scales=[8],
+        anchor_ratios=[0.5, 1.0, 2.0],
+        anchor_strides=[4, 8, 16, 32, 64],
+        target_means=[.0, .0, .0, .0],
+        target_stds=[1.0, 1.0, 1.0, 1.0],
+        use_sigmoid_cls=True),
+    bbox_roi_extractor=dict(
+        type='SingleRoIExtractor',
+        roi_layer=dict(
+            type='ModulatedDeformRoIPoolingPack',
+            out_size=7,
+            out_channels=256,
+            no_trans=False,
+            group_size=1,
+            trans_std=0.1),
+        out_channels=256,
+        featmap_strides=[4, 8, 16, 32]),
+    bbox_head=dict(
+        type='SharedFCBBoxHead',
+        num_fcs=2,
+        in_channels=256,
+        fc_out_channels=1024,
+        roi_feat_size=7,
+        num_classes=81,
+        target_means=[0., 0., 0., 0.],
+        target_stds=[0.1, 0.1, 0.2, 0.2],
+        reg_class_agnostic=False))
+# model training and testing settings
+train_cfg = dict(
+    rpn=dict(
+        assigner=dict(
+            type='MaxIoUAssigner',
+            pos_iou_thr=0.7,
+            neg_iou_thr=0.3,
+            min_pos_iou=0.3,
+            ignore_iof_thr=-1),
+        sampler=dict(
+            type='RandomSampler',
+            num=256,
+            pos_fraction=0.5,
+            neg_pos_ub=-1,
+            add_gt_as_proposals=False),
+        allowed_border=0,
+        pos_weight=-1,
+        smoothl1_beta=1 / 9.0,
+        debug=False),
+    rcnn=dict(
+        assigner=dict(
+            type='MaxIoUAssigner',
+            pos_iou_thr=0.5,
+            neg_iou_thr=0.5,
+            min_pos_iou=0.5,
+            ignore_iof_thr=-1),
+        sampler=dict(
+            type='RandomSampler',
+            num=512,
+            pos_fraction=0.25,
+            neg_pos_ub=-1,
+            add_gt_as_proposals=True),
+        pos_weight=-1,
+        debug=False))
+test_cfg = dict(
+    rpn=dict(
+        nms_across_levels=False,
+        nms_pre=2000,
+        nms_post=2000,
+        max_num=2000,
+        nms_thr=0.7,
+        min_bbox_size=0),
+    rcnn=dict(
+        score_thr=0.05, nms=dict(type='nms', iou_thr=0.5), max_per_img=100)
+    # soft-nms is also supported for rcnn testing
+    # e.g., nms=dict(type='soft_nms', iou_thr=0.5, min_score=0.05)
+)
+# dataset settings
+dataset_type = 'CocoDataset'
+data_root = 'data/coco/'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+data = dict(
+    imgs_per_gpu=2,
+    workers_per_gpu=2,
+    train=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_train2017.json',
+        img_prefix=data_root + 'train2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0.5,
+        with_mask=False,
+        with_crowd=True,
+        with_label=True),
+    val=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0,
+        with_mask=False,
+        with_crowd=True,
+        with_label=True),
+    test=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0,
+        with_mask=False,
+        with_label=False,
+        test_mode=True))
+# optimizer
+optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=1.0 / 3,
+    step=[8, 11])
+checkpoint_config = dict(interval=1)
+# yapf:disable
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        # dict(type='TensorboardLoggerHook')
+    ])
+# yapf:enable
+# runtime settings
+total_epochs = 12
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+work_dir = './work_dirs/faster_rcnn_mdpool_r50_fpn_1x'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]

configs/dcn/mask_rcnn_dconv_c3-c5_r50_fpn_1x.py

+# model settings
+model = dict(
+    type='MaskRCNN',
+    pretrained='modelzoo://resnet50',
+    backbone=dict(
+        type='ResNet',
+        depth=50,
+        num_stages=4,
+        out_indices=(0, 1, 2, 3),
+        frozen_stages=1,
+        style='pytorch',
+        dcn=dict(
+            modulated=False,
+            deformable_groups=1,
+            fallback_on_stride=False),
+        stage_with_dcn=(False, True, True, True)),
+    neck=dict(
+        type='FPN',
+        in_channels=[256, 512, 1024, 2048],
+        out_channels=256,
+        num_outs=5),
+    rpn_head=dict(
+        type='RPNHead',
+        in_channels=256,
+        feat_channels=256,
+        anchor_scales=[8],
+        anchor_ratios=[0.5, 1.0, 2.0],
+        anchor_strides=[4, 8, 16, 32, 64],
+        target_means=[.0, .0, .0, .0],
+        target_stds=[1.0, 1.0, 1.0, 1.0],
+        use_sigmoid_cls=True),
+    bbox_roi_extractor=dict(
+        type='SingleRoIExtractor',
+        roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),
+        out_channels=256,
+        featmap_strides=[4, 8, 16, 32]),
+    bbox_head=dict(
+        type='SharedFCBBoxHead',
+        num_fcs=2,
+        in_channels=256,
+        fc_out_channels=1024,
+        roi_feat_size=7,
+        num_classes=81,
+        target_means=[0., 0., 0., 0.],
+        target_stds=[0.1, 0.1, 0.2, 0.2],
+        reg_class_agnostic=False),
+    mask_roi_extractor=dict(
+        type='SingleRoIExtractor',
+        roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
+        out_channels=256,
+        featmap_strides=[4, 8, 16, 32]),
+    mask_head=dict(
+        type='FCNMaskHead',
+        num_convs=4,
+        in_channels=256,
+        conv_out_channels=256,
+        num_classes=81))
+# model training and testing settings
+train_cfg = dict(
+    rpn=dict(
+        assigner=dict(
+            type='MaxIoUAssigner',
+            pos_iou_thr=0.7,
+            neg_iou_thr=0.3,
+            min_pos_iou=0.3,
+            ignore_iof_thr=-1),
+        sampler=dict(
+            type='RandomSampler',
+            num=256,
+            pos_fraction=0.5,
+            neg_pos_ub=-1,
+            add_gt_as_proposals=False),
+        allowed_border=0,
+        pos_weight=-1,
+        smoothl1_beta=1 / 9.0,
+        debug=False),
+    rcnn=dict(
+        assigner=dict(
+            type='MaxIoUAssigner',
+            pos_iou_thr=0.5,
+            neg_iou_thr=0.5,
+            min_pos_iou=0.5,
+            ignore_iof_thr=-1),
+        sampler=dict(
+            type='RandomSampler',
+            num=512,
+            pos_fraction=0.25,
+            neg_pos_ub=-1,
+            add_gt_as_proposals=True),
+        mask_size=28,
+        pos_weight=-1,
+        debug=False))
+test_cfg = dict(
+    rpn=dict(
+        nms_across_levels=False,
+        nms_pre=2000,
+        nms_post=2000,
+        max_num=2000,
+        nms_thr=0.7,
+        min_bbox_size=0),
+    rcnn=dict(
+        score_thr=0.05,
+        nms=dict(type='nms', iou_thr=0.5),
+        max_per_img=100,
+        mask_thr_binary=0.5))
+# dataset settings
+dataset_type = 'CocoDataset'
+data_root = 'data/coco/'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+data = dict(
+    imgs_per_gpu=2,
+    workers_per_gpu=2,
+    train=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_train2017.json',
+        img_prefix=data_root + 'train2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0.5,
+        with_mask=True,
+        with_crowd=True,
+        with_label=True),
+    val=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0,
+        with_mask=True,
+        with_crowd=True,
+        with_label=True),
+    test=dict(
+        type=dataset_type,
+        ann_file=data_root + 'annotations/instances_val2017.json',
+        img_prefix=data_root + 'val2017/',
+        img_scale=(1333, 800),
+        img_norm_cfg=img_norm_cfg,
+        size_divisor=32,
+        flip_ratio=0,
+        with_mask=False,
+        with_label=False,
+        test_mode=True))
+# optimizer
+optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=1.0 / 3,
+    step=[8, 11])
+checkpoint_config = dict(interval=1)
+# yapf:disable
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        # dict(type='TensorboardLoggerHook')
+    ])
+# yapf:enable
+# runtime settings
+total_epochs = 12
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+work_dir = './work_dirs/mask_rcnn_dconv_c3-c5_r50_fpn_1x'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]

mmdet/models/backbones/resnet.py

@@ -5,9 +5,10 @@ import torch.utils.checkpoint as cp
 
 from mmcv.cnn import constant_init, kaiming_init
 from mmcv.runner import load_checkpoint
-from ..utils import build_norm_layer
 
+from mmdet.ops import DeformConv, ModulatedDeformConv
 from ..registry import BACKBONES
+from ..utils import build_norm_layer
 
 
 def conv3x3(in_planes, out_planes, stride=1, dilation=1):
@@ -88,16 +89,20 @@ class Bottleneck(nn.Module):
                  downsample=None,
                  style='pytorch',
                  with_cp=False,
-                 normalize=dict(type='BN')):
+                 normalize=dict(type='BN'),
+                 dcn=None):
         """Bottleneck block for ResNet.
         If style is "pytorch", the stride-two layer is the 3x3 conv layer,
         if it is "caffe", the stride-two layer is the first 1x1 conv layer.
         """
         super(Bottleneck, self).__init__()
         assert style in ['pytorch', 'caffe']
+        assert dcn is None or isinstance(dcn, dict)
         self.inplanes = inplanes
         self.planes = planes
         self.normalize = normalize
+        self.dcn = dcn
+        self.with_dcn = dcn is not None
         if style == 'pytorch':
             self.conv1_stride = 1
             self.conv2_stride = stride
@@ -107,9 +112,8 @@ class Bottleneck(nn.Module):
 
         self.norm1_name, norm1 = build_norm_layer(normalize, planes, postfix=1)
         self.norm2_name, norm2 = build_norm_layer(normalize, planes, postfix=2)
-        self.norm3_name, norm3 = build_norm_layer(normalize,
-                                                  planes * self.expansion,
-                                                  postfix=3)
+        self.norm3_name, norm3 = build_norm_layer(
+            normalize, planes * self.expansion, postfix=3)
 
         self.conv1 = nn.Conv2d(
             inplanes,
@@ -118,14 +122,44 @@ class Bottleneck(nn.Module):
             stride=self.conv1_stride,
             bias=False)
         self.add_module(self.norm1_name, norm1)
-        self.conv2 = nn.Conv2d(
-            planes,
-            planes,
-            kernel_size=3,
-            stride=self.conv2_stride,
-            padding=dilation,
-            dilation=dilation,
-            bias=False)
+        fallback_on_stride = False
+        self.with_modulated_dcn = False
+        if self.with_dcn:
+            fallback_on_stride = dcn.get('fallback_on_stride', False)
+            self.with_modulated_dcn = dcn.get('modulated', False)
+        if not self.with_dcn or fallback_on_stride:
+            self.conv2 = nn.Conv2d(
+                planes,
+                planes,
+                kernel_size=3,
+                stride=self.conv2_stride,
+                padding=dilation,
+                dilation=dilation,
+                bias=False)
+        else:
+            deformable_groups = dcn.get('deformable_groups', 1)
+            if not self.with_modulated_dcn:
+                conv_op = DeformConv
+                offset_channels = 18
+            else:
+                conv_op = ModulatedDeformConv
+                offset_channels = 27
+            self.conv2_offset = nn.Conv2d(
+                planes,
+                deformable_groups * offset_channels,
+                kernel_size=3,
+                stride=self.conv2_stride,
+                padding=dilation,
+                dilation=dilation)
+            self.conv2 = conv_op(
+                planes,
+                planes,
+                kernel_size=3,
+                stride=self.conv2_stride,
+                padding=dilation,
+                dilation=dilation,
+                deformable_groups=deformable_groups,
+                bias=False)
         self.add_module(self.norm2_name, norm2)
         self.conv3 = nn.Conv2d(
             planes, planes * self.expansion, kernel_size=1, bias=False)
@@ -159,7 +193,16 @@ class Bottleneck(nn.Module):
             out = self.norm1(out)
             out = self.relu(out)
 
-            out = self.conv2(out)
+            if not self.with_dcn:
+                out = self.conv2(out)
+            elif self.with_modulated_dcn:
+                offset_mask = self.conv2_offset(out)
+                offset = offset_mask[:, :18, :, :]
+                mask = offset_mask[:, -9:, :, :].sigmoid()
+                out = self.conv2(out, offset, mask)
+            else:
+                offset = self.conv2_offset(out)
+                out = self.conv2(out, offset)
             out = self.norm2(out)
             out = self.relu(out)
 
@@ -191,7 +234,8 @@ def make_res_layer(block,
                    dilation=1,
                    style='pytorch',
                    with_cp=False,
-                   normalize=dict(type='BN')):
+                   normalize=dict(type='BN'),
+                   dcn=None):
     downsample = None
     if stride != 1 or inplanes != planes * block.expansion:
         downsample = nn.Sequential(
@@ -214,12 +258,20 @@ def make_res_layer(block,
             downsample,
             style=style,
             with_cp=with_cp,
-            normalize=normalize))
+            normalize=normalize,
+            dcn=dcn))
     inplanes = planes * block.expansion
     for i in range(1, blocks):
         layers.append(
-            block(inplanes, planes, 1, dilation, style=style,
-                  with_cp=with_cp, normalize=normalize))
+            block(
+                inplanes,
+                planes,
+                1,
+                dilation,
+                style=style,
+                with_cp=with_cp,
+                normalize=normalize,
+                dcn=dcn))
 
     return nn.Sequential(*layers)
 
@@ -265,10 +317,10 @@ class ResNet(nn.Module):
                  out_indices=(0, 1, 2, 3),
                  style='pytorch',
                  frozen_stages=-1,
-                 normalize=dict(
-                     type='BN',
-                     frozen=False),
+                 normalize=dict(type='BN', frozen=False),
                  norm_eval=True,
+                 dcn=None,
+                 stage_with_dcn=(False, False, False, False),
                  with_cp=False,
                  zero_init_residual=True):
         super(ResNet, self).__init__()
@@ -279,7 +331,8 @@ class ResNet(nn.Module):
         assert num_stages >= 1 and num_stages <= 4
         self.strides = strides
         self.dilations = dilations
-        assert len(strides) == len(dilations) == num_stages
+        assert len(strides) == len(dilations) == len(
+            stage_with_dcn) == num_stages
         self.out_indices = out_indices
         assert max(out_indices) < num_stages
         self.style = style
@@ -287,6 +340,8 @@ class ResNet(nn.Module):
         self.normalize = normalize
         self.with_cp = with_cp
         self.norm_eval = norm_eval
+        self.dcn = dcn
+        self.stage_with_dcn = stage_with_dcn
         self.zero_init_residual = zero_init_residual
         self.block, stage_blocks = self.arch_settings[depth]
         self.stage_blocks = stage_blocks[:num_stages]
@@ -298,6 +353,7 @@ class ResNet(nn.Module):
         for i, num_blocks in enumerate(self.stage_blocks):
             stride = strides[i]
             dilation = dilations[i]
+            dcn = self.dcn if self.stage_with_dcn[i] else None
             planes = 64 * 2**i
             res_layer = make_res_layer(
                 self.block,
@@ -308,7 +364,8 @@ class ResNet(nn.Module):
                 dilation=dilation,
                 style=self.style,
                 with_cp=with_cp,
-                normalize=normalize)
+                normalize=normalize,
+                dcn=dcn)
             self.inplanes = planes * self.block.expansion
             layer_name = 'layer{}'.format(i + 1)
             self.add_module(layer_name, res_layer)
@@ -326,8 +383,8 @@ class ResNet(nn.Module):
     def _make_stem_layer(self):
         self.conv1 = nn.Conv2d(
             3, 64, kernel_size=7, stride=2, padding=3, bias=False)
-        self.norm1_name, norm1 = build_norm_layer(self.normalize,
-                                                  64, postfix=1)
+        self.norm1_name, norm1 = build_norm_layer(
+            self.normalize, 64, postfix=1)
         self.add_module(self.norm1_name, norm1)
         self.relu = nn.ReLU(inplace=True)
         self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
@@ -354,6 +411,12 @@ class ResNet(nn.Module):
                 elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
                     constant_init(m, 1)
 
+            if self.dcn is not None:
+                for m in self.modules():
+                    if isinstance(m, Bottleneck) and hasattr(
+                            m, 'conv2_offset'):
+                        constant_init(m.conv2_offset, 0)
+
             if self.zero_init_residual:
                 for m in self.modules():
                     if isinstance(m, Bottleneck):

mmdet/ops/__init__.py

+from .dcn import (DeformConv, DeformRoIPooling, DeformRoIPoolingPack,
+                  ModulatedDeformRoIPoolingPack, ModulatedDeformConv,
+                  ModulatedDeformConvPack, deform_conv, modulated_deform_conv,
+                  deform_roi_pooling)
 from .nms import nms, soft_nms
 from .roi_align import RoIAlign, roi_align
 from .roi_pool import RoIPool, roi_pool
 
-__all__ = ['nms', 'soft_nms', 'RoIAlign', 'roi_align', 'RoIPool', 'roi_pool']
+__all__ = [
+    'nms', 'soft_nms', 'RoIAlign', 'roi_align', 'RoIPool', 'roi_pool',
+    'DeformConv', 'DeformRoIPooling', 'DeformRoIPoolingPack',
+    'ModulatedDeformRoIPoolingPack', 'ModulatedDeformConv',
+    'ModulatedDeformConvPack', 'deform_conv', 'modulated_deform_conv',
+    'deform_roi_pooling'
+]

mmdet/ops/dcn/__init__.py

+from .functions.deform_conv import deform_conv, modulated_deform_conv
+from .functions.deform_pool import deform_roi_pooling
+from .modules.deform_conv import (DeformConv, ModulatedDeformConv,
+                                  ModulatedDeformConvPack)
+from .modules.deform_pool import (DeformRoIPooling, DeformRoIPoolingPack,
+                                  ModulatedDeformRoIPoolingPack)
+
+__all__ = [
+    'DeformConv', 'DeformRoIPooling', 'DeformRoIPoolingPack',
+    'ModulatedDeformRoIPoolingPack', 'ModulatedDeformConv',
+    'ModulatedDeformConvPack', 'deform_conv',
+    'modulated_deform_conv', 'deform_roi_pooling'
+]

mmdet/ops/dcn/functions/deform_conv.py

+import torch
+from torch.autograd import Function
+from torch.nn.modules.utils import _pair
+
+from .. import deform_conv_cuda
+
+
+class DeformConvFunction(Function):
+
+    @staticmethod
+    def forward(ctx,
+                input,
+                offset,
+                weight,
+                stride=1,
+                padding=0,
+                dilation=1,
+                deformable_groups=1,
+                im2col_step=64):
+        if input is not None and input.dim() != 4:
+            raise ValueError(
+                "Expected 4D tensor as input, got {}D tensor instead.".format(
+                    input.dim()))
+        ctx.stride = _pair(stride)
+        ctx.padding = _pair(padding)
+        ctx.dilation = _pair(dilation)
+        ctx.deformable_groups = deformable_groups
+        ctx.im2col_step = im2col_step
+
+        ctx.save_for_backward(input, offset, weight)
+
+        output = input.new_empty(
+            DeformConvFunction._output_size(input, weight, ctx.padding,
+                                            ctx.dilation, ctx.stride))
+
+        ctx.bufs_ = [input.new_empty(0), input.new_empty(0)]  # columns, ones
+
+        if not input.is_cuda:
+            raise NotImplementedError
+        else:
+            cur_im2col_step = min(ctx.im2col_step, input.shape[0])
+            assert (input.shape[0] %
+                    cur_im2col_step) == 0, 'im2col step must divide batchsize'
+            deform_conv_cuda.deform_conv_forward_cuda(
+                input, weight, offset, output, ctx.bufs_[0], ctx.bufs_[1],
+                weight.size(3), weight.size(2), ctx.stride[1], ctx.stride[0],
+                ctx.padding[1], ctx.padding[0], ctx.dilation[1],
+                ctx.dilation[0], ctx.deformable_groups, cur_im2col_step)
+        return output
+
+    @staticmethod
+    def backward(ctx, grad_output):
+        input, offset, weight = ctx.saved_tensors
+
+        grad_input = grad_offset = grad_weight = None
+
+        if not grad_output.is_cuda:
+            raise NotImplementedError
+        else:
+            cur_im2col_step = min(ctx.im2col_step, input.shape[0])
+            assert (input.shape[0] %
+                    cur_im2col_step) == 0, 'im2col step must divide batchsize'
+
+            if ctx.needs_input_grad[0] or ctx.needs_input_grad[1]:
+                grad_input = torch.zeros_like(input)
+                grad_offset = torch.zeros_like(offset)
+                deform_conv_cuda.deform_conv_backward_input_cuda(
+                    input, offset, grad_output, grad_input,
+                    grad_offset, weight, ctx.bufs_[0], weight.size(3),
+                    weight.size(2), ctx.stride[1], ctx.stride[0],
+                    ctx.padding[1], ctx.padding[0], ctx.dilation[1],
+                    ctx.dilation[0], ctx.deformable_groups, cur_im2col_step)
+
+            if ctx.needs_input_grad[2]:
+                grad_weight = torch.zeros_like(weight)
+                deform_conv_cuda.deform_conv_backward_parameters_cuda(
+                    input, offset, grad_output,
+                    grad_weight, ctx.bufs_[0], ctx.bufs_[1], weight.size(3),
+                    weight.size(2), ctx.stride[1], ctx.stride[0],
+                    ctx.padding[1], ctx.padding[0], ctx.dilation[1],
+                    ctx.dilation[0], ctx.deformable_groups, 1, cur_im2col_step)
+
+        return grad_input, grad_offset, grad_weight, None, None, None, None
+
+    @staticmethod
+    def _output_size(input, weight, padding, dilation, stride):
+        channels = weight.size(0)
+        output_size = (input.size(0), channels)
+        for d in range(input.dim() - 2):
+            in_size = input.size(d + 2)
+            pad = padding[d]
+            kernel = dilation[d] * (weight.size(d + 2) - 1) + 1
+            stride_ = stride[d]
+            output_size += ((in_size + (2 * pad) - kernel) // stride_ + 1, )
+        if not all(map(lambda s: s > 0, output_size)):
+            raise ValueError(
+                "convolution input is too small (output would be {})".format(
+                    'x'.join(map(str, output_size))))
+        return output_size
+
+
+class ModulatedDeformConvFunction(Function):
+
+    @staticmethod
+    def forward(ctx,
+                input,
+                offset,
+                mask,
+                weight,
+                bias=None,
+                stride=1,
+                padding=0,
+                dilation=1,
+                deformable_groups=1):
+        ctx.stride = stride
+        ctx.padding = padding
+        ctx.dilation = dilation
+        ctx.deformable_groups = deformable_groups
+        ctx.with_bias = bias is not None
+        if not ctx.with_bias:
+            bias = input.new_empty(1)  # fake tensor
+        if not input.is_cuda:
+            raise NotImplementedError
+        if weight.requires_grad or mask.requires_grad or offset.requires_grad \
+                or input.requires_grad:
+            ctx.save_for_backward(input, offset, mask, weight, bias)
+        output = input.new_empty(
+            ModulatedDeformConvFunction._infer_shape(ctx, input, weight))
+        ctx._bufs = [input.new_empty(0), input.new_empty(0)]
+        deform_conv_cuda.modulated_deform_conv_cuda_forward(
+            input, weight, bias, ctx._bufs[0], offset, mask, output,
+            ctx._bufs[1], weight.shape[2], weight.shape[3], ctx.stride,
+            ctx.stride, ctx.padding, ctx.padding, ctx.dilation, ctx.dilation,
+            ctx.deformable_groups, ctx.with_bias)
+        return output
+
+    @staticmethod
+    def backward(ctx, grad_output):
+        if not grad_output.is_cuda:
+            raise NotImplementedError
+        input, offset, mask, weight, bias = ctx.saved_tensors
+        grad_input = torch.zeros_like(input)
+        grad_offset = torch.zeros_like(offset)
+        grad_mask = torch.zeros_like(mask)
+        grad_weight = torch.zeros_like(weight)
+        grad_bias = torch.zeros_like(bias)
+        deform_conv_cuda.modulated_deform_conv_cuda_backward(
+            input, weight, bias, ctx._bufs[0], offset, mask, ctx._bufs[1],
+            grad_input, grad_weight, grad_bias, grad_offset, grad_mask,
+            grad_output, weight.shape[2], weight.shape[3], ctx.stride,
+            ctx.stride, ctx.padding, ctx.padding, ctx.dilation, ctx.dilation,
+            ctx.deformable_groups, ctx.with_bias)
+        if not ctx.with_bias:
+            grad_bias = None
+
+        return (grad_input, grad_offset, grad_mask, grad_weight, grad_bias,
+                None, None, None, None)
+
+    @staticmethod
+    def _infer_shape(ctx, input, weight):
+        n = input.size(0)
+        channels_out = weight.size(0)
+        height, width = input.shape[2:4]
+        kernel_h, kernel_w = weight.shape[2:4]
+        height_out = (height + 2 * ctx.padding -
+                      (ctx.dilation * (kernel_h - 1) + 1)) // ctx.stride + 1
+        width_out = (width + 2 * ctx.padding -
+                     (ctx.dilation * (kernel_w - 1) + 1)) // ctx.stride + 1
+        return n, channels_out, height_out, width_out
+
+
+deform_conv = DeformConvFunction.apply
+modulated_deform_conv = ModulatedDeformConvFunction.apply

mmdet/ops/dcn/functions/deform_pool.py

+import torch
+from torch.autograd import Function
+
+from .. import deform_pool_cuda
+
+
+class DeformRoIPoolingFunction(Function):
+
+    @staticmethod
+    def forward(ctx,
+                data,
+                rois,
+                offset,
+                spatial_scale,
+                out_size,
+                out_channels,
+                no_trans,
+                group_size=1,
+                part_size=None,
+                sample_per_part=4,
+                trans_std=.0):
+        ctx.spatial_scale = spatial_scale
+        ctx.out_size = out_size
+        ctx.out_channels = out_channels
+        ctx.no_trans = no_trans
+        ctx.group_size = group_size
+        ctx.part_size = out_size if part_size is None else part_size
+        ctx.sample_per_part = sample_per_part
+        ctx.trans_std = trans_std
+
+        assert 0.0 <= ctx.trans_std <= 1.0
+        if not data.is_cuda:
+            raise NotImplementedError
+
+        n = rois.shape[0]
+        output = data.new_empty(n, out_channels, out_size, out_size)
+        output_count = data.new_empty(n, out_channels, out_size, out_size)
+        deform_pool_cuda.deform_psroi_pooling_cuda_forward(
+            data, rois, offset, output, output_count, ctx.no_trans,
+            ctx.spatial_scale, ctx.out_channels, ctx.group_size, ctx.out_size,
+            ctx.part_size, ctx.sample_per_part, ctx.trans_std)
+
+        if data.requires_grad or rois.requires_grad or offset.requires_grad:
+            ctx.save_for_backward(data, rois, offset)
+        ctx.output_count = output_count
+
+        return output
+
+    @staticmethod
+    def backward(ctx, grad_output):
+        if not grad_output.is_cuda:
+            raise NotImplementedError
+
+        data, rois, offset = ctx.saved_tensors
+        output_count = ctx.output_count
+        grad_input = torch.zeros_like(data)
+        grad_rois = None
+        grad_offset = torch.zeros_like(offset)
+
+        deform_pool_cuda.deform_psroi_pooling_cuda_backward(
+            grad_output, data, rois, offset, output_count, grad_input,
+            grad_offset, ctx.no_trans, ctx.spatial_scale, ctx.out_channels,
+            ctx.group_size, ctx.out_size, ctx.part_size, ctx.sample_per_part,
+            ctx.trans_std)
+        return (grad_input, grad_rois, grad_offset, None, None, None, None,
+                None, None, None, None)
+
+
+deform_roi_pooling = DeformRoIPoolingFunction.apply

mmdet/ops/dcn/modules/deform_conv.py

+import math
+
+import torch
+import torch.nn as nn
+from torch.nn.modules.utils import _pair
+
+from ..functions.deform_conv import deform_conv, modulated_deform_conv
+
+
+class DeformConv(nn.Module):
+
+    def __init__(self,
+                 in_channels,
+                 out_channels,
+                 kernel_size,
+                 stride=1,
+                 padding=0,
+                 dilation=1,
+                 deformable_groups=1,
+                 bias=False):
+        assert not bias
+        super(DeformConv, self).__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.kernel_size = _pair(kernel_size)
+        self.stride = _pair(stride)
+        self.padding = _pair(padding)
+        self.dilation = _pair(dilation)
+        self.deformable_groups = deformable_groups
+
+        self.weight = nn.Parameter(
+            torch.Tensor(out_channels, in_channels, *self.kernel_size))
+
+        self.reset_parameters()
+
+    def reset_parameters(self):
+        n = self.in_channels
+        for k in self.kernel_size:
+            n *= k
+        stdv = 1. / math.sqrt(n)
+        self.weight.data.uniform_(-stdv, stdv)
+
+    def forward(self, input, offset):
+        return deform_conv(input, offset, self.weight, self.stride,
+                           self.padding, self.dilation, self.deformable_groups)
+
+
+class ModulatedDeformConv(nn.Module):
+
+    def __init__(self,
+                 in_channels,
+                 out_channels,
+                 kernel_size,
+                 stride=1,
+                 padding=0,
+                 dilation=1,
+                 deformable_groups=1,
+                 bias=True):
+        super(ModulatedDeformConv, self).__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.kernel_size = _pair(kernel_size)
+        self.stride = stride
+        self.padding = padding
+        self.dilation = dilation
+        self.deformable_groups = deformable_groups
+        self.with_bias = bias
+
+        self.weight = nn.Parameter(
+            torch.Tensor(out_channels, in_channels, *self.kernel_size))
+        if bias:
+            self.bias = nn.Parameter(torch.Tensor(out_channels))
+        else:
+            self.register_parameter('bias', None)
+        self.reset_parameters()
+
+    def reset_parameters(self):
+        n = self.in_channels
+        for k in self.kernel_size:
+            n *= k
+        stdv = 1. / math.sqrt(n)
+        self.weight.data.uniform_(-stdv, stdv)
+        if self.bias is not None:
+            self.bias.data.zero_()
+
+    def forward(self, input, offset, mask):
+        return modulated_deform_conv(input, offset, mask, self.weight,
+                                     self.bias, self.stride, self.padding,
+                                     self.dilation, self.deformable_groups)
+
+
+class ModulatedDeformConvPack(ModulatedDeformConv):
+
+    def __init__(self,
+                 in_channels,
+                 out_channels,
+                 kernel_size,
+                 stride=1,
+                 padding=0,
+                 dilation=1,
+                 deformable_groups=1,
+                 bias=True):
+        super(ModulatedDeformConvPack,
+              self).__init__(in_channels, out_channels, kernel_size, stride,
+                             padding, dilation, deformable_groups, bias)
+
+        self.conv_offset_mask = nn.Conv2d(
+            self.in_channels,
+            self.deformable_groups * 3 * self.kernel_size[0] *
+            self.kernel_size[1],
+            kernel_size=self.kernel_size,
+            stride=_pair(self.stride),
+            padding=_pair(self.padding),
+            bias=True)
+        self.init_offset()
+
+    def init_offset(self):
+        self.conv_offset_mask.weight.data.zero_()
+        self.conv_offset_mask.bias.data.zero_()
+
+    def forward(self, input):
+        out = self.conv_offset_mask(input)
+        o1, o2, mask = torch.chunk(out, 3, dim=1)
+        offset = torch.cat((o1, o2), dim=1)
+        mask = torch.sigmoid(mask)
+        return modulated_deform_conv(input, offset, mask, self.weight,
+                                     self.bias, self.stride, self.padding,
+                                     self.dilation, self.deformable_groups)

mmdet/ops/dcn/modules/deform_pool.py

+from torch import nn
+
+from ..functions.deform_pool import deform_roi_pooling
+
+
+class DeformRoIPooling(nn.Module):
+
+    def __init__(self,
+                 spatial_scale,
+                 out_size,
+                 out_channels,
+                 no_trans,
+                 group_size=1,
+                 part_size=None,
+                 sample_per_part=4,
+                 trans_std=.0):
+        super(DeformRoIPooling, self).__init__()
+        self.spatial_scale = spatial_scale
+        self.out_size = out_size
+        self.out_channels = out_channels
+        self.no_trans = no_trans
+        self.group_size = group_size
+        self.part_size = out_size if part_size is None else part_size
+        self.sample_per_part = sample_per_part
+        self.trans_std = trans_std
+
+    def forward(self, data, rois, offset):
+        if self.no_trans:
+            offset = data.new_empty(0)
+        return deform_roi_pooling(
+            data, rois, offset, self.spatial_scale, self.out_size,
+            self.out_channels, self.no_trans, self.group_size, self.part_size,
+            self.sample_per_part, self.trans_std)
+
+
+class DeformRoIPoolingPack(DeformRoIPooling):
+
+    def __init__(self,
+                 spatial_scale,
+                 out_size,
+                 out_channels,
+                 no_trans,
+                 group_size=1,
+                 part_size=None,
+                 sample_per_part=4,
+                 trans_std=.0,
+                 deform_fc_channels=1024):
+        super(DeformRoIPoolingPack,
+              self).__init__(spatial_scale, out_size, out_channels, no_trans,
+                             group_size, part_size, sample_per_part, trans_std)
+
+        self.deform_fc_channels = deform_fc_channels
+
+        if not no_trans:
+            self.offset_fc = nn.Sequential(
+                nn.Linear(self.out_size * self.out_size * self.out_channels,
+                          self.deform_fc_channels),
+                nn.ReLU(inplace=True),
+                nn.Linear(self.deform_fc_channels, self.deform_fc_channels),
+                nn.ReLU(inplace=True),
+                nn.Linear(self.deform_fc_channels,
+                          self.out_size * self.out_size * 2))
+            self.offset_fc[-1].weight.data.zero_()
+            self.offset_fc[-1].bias.data.zero_()
+
+    def forward(self, data, rois):
+        assert data.size(1) == self.out_channels
+        if self.no_trans:
+            offset = data.new_empty(0)
+            return deform_roi_pooling(
+                data, rois, offset, self.spatial_scale, self.out_size,
+                self.out_channels, self.no_trans, self.group_size,
+                self.part_size, self.sample_per_part, self.trans_std)
+        else:
+            n = rois.shape[0]
+            offset = data.new_empty(0)
+            x = deform_roi_pooling(data, rois, offset, self.spatial_scale,
+                                   self.out_size, self.out_channels, True,
+                                   self.group_size, self.part_size,
+                                   self.sample_per_part, self.trans_std)
+            offset = self.offset_fc(x.view(n, -1))
+            offset = offset.view(n, 2, self.out_size, self.out_size)
+            return deform_roi_pooling(
+                data, rois, offset, self.spatial_scale, self.out_size,
+                self.out_channels, self.no_trans, self.group_size,
+                self.part_size, self.sample_per_part, self.trans_std)
+
+
+class ModulatedDeformRoIPoolingPack(DeformRoIPooling):
+
+    def __init__(self,
+                 spatial_scale,
+                 out_size,
+                 out_channels,
+                 no_trans,
+                 group_size=1,
+                 part_size=None,
+                 sample_per_part=4,
+                 trans_std=.0,
+                 deform_fc_channels=1024):
+        super(ModulatedDeformRoIPoolingPack, self).__init__(
+            spatial_scale, out_size, out_channels, no_trans, group_size,
+            part_size, sample_per_part, trans_std)
+
+        self.deform_fc_channels = deform_fc_channels
+
+        if not no_trans:
+            self.offset_fc = nn.Sequential(
+                nn.Linear(self.out_size * self.out_size * self.out_channels,
+                          self.deform_fc_channels),
+                nn.ReLU(inplace=True),
+                nn.Linear(self.deform_fc_channels, self.deform_fc_channels),
+                nn.ReLU(inplace=True),
+                nn.Linear(self.deform_fc_channels,
+                          self.out_size * self.out_size * 2))
+            self.offset_fc[-1].weight.data.zero_()
+            self.offset_fc[-1].bias.data.zero_()
+            self.mask_fc = nn.Sequential(
+                nn.Linear(self.out_size * self.out_size * self.out_channels,
+                          self.deform_fc_channels),
+                nn.ReLU(inplace=True),
+                nn.Linear(self.deform_fc_channels,
+                          self.out_size * self.out_size * 1),
+                nn.Sigmoid())
+            self.mask_fc[2].weight.data.zero_()
+            self.mask_fc[2].bias.data.zero_()
+
+    def forward(self, data, rois):
+        assert data.size(1) == self.out_channels
+        if self.no_trans:
+            offset = data.new_empty(0)
+            return deform_roi_pooling(
+                data, rois, offset, self.spatial_scale, self.out_size,
+                self.out_channels, self.no_trans, self.group_size,
+                self.part_size, self.sample_per_part, self.trans_std)
+        else:
+            n = rois.shape[0]
+            offset = data.new_empty(0)
+            x = deform_roi_pooling(data, rois, offset, self.spatial_scale,
+                                   self.out_size, self.out_channels, True,
+                                   self.group_size, self.part_size,
+                                   self.sample_per_part, self.trans_std)
+            offset = self.offset_fc(x.view(n, -1))
+            offset = offset.view(n, 2, self.out_size, self.out_size)
+            mask = self.mask_fc(x.view(n, -1))
+            mask = mask.view(n, 1, self.out_size, self.out_size)
+            return deform_roi_pooling(
+                data, rois, offset, self.spatial_scale, self.out_size,
+                self.out_channels, self.no_trans, self.group_size,
+                self.part_size, self.sample_per_part, self.trans_std) * mask

mmdet/ops/dcn/setup.py

+from setuptools import setup
+from torch.utils.cpp_extension import BuildExtension, CUDAExtension
+
+setup(
+    name='deform_conv',
+    ext_modules=[
+        CUDAExtension('deform_conv_cuda', [
+            'src/deform_conv_cuda.cpp',
+            'src/deform_conv_cuda_kernel.cu',
+        ]),
+        CUDAExtension('deform_pool_cuda', [
+            'src/deform_pool_cuda.cpp', 'src/deform_pool_cuda_kernel.cu'
+        ]),
+    ],
+    cmdclass={'build_ext': BuildExtension})