[Refactor]: Support batch inference traceable by ONNX in SSD, YOLOv3, FSAF, RetinaNet, and FCOS (#4699)

* Support batch infer in RetinaNet

* Suport batch multiclass_nms

* Revert multiclass_nms

* Fix api deprecated warning

* do not repeat anchors

* Move img_shapes

* Update Yolov3

* Support FCOS

* Support RPN

* Fix RPN topk_inds error

* make batch exportable to onnx for yolohead

* make fcos_head exportable to onnx with batch dim

* Support ATSS

* Support CornerNet and centripetalNet

* Update RetinaNet and delta_xywh

* Remove ugly code

* Remove ugly code of FCOS

* Remove ugly code of ATSS/YOLOV3

* Support RPN and revert bbox_head

* expand anchors to batch and remove BG class when use deploy_nms_pre

* Update

* Use dim=-1 instead of dim=2

* Rename anchor_head method

* Keep the original format output when nms is not use

* Rename method and unified code style

* Fix paa_head and unittest

* Fix FASF onnx export error

* Fix error

* fix single stage img_shapes for onnx

* move conf_thr

* fix rpn_head for onnx

* Add distance2bbox unittest

* Remove TODO

* Fix RPN

* Update docstrs

Co-authored-by: maningsheng <maningsheng@sensetime.com>

* Fix RPN show_result error

* Remove EmbeddingRPN show_result

* Add docstr

* add dynamic shape support for two_stage func

* update code

* update code

* update code

* bbox coder support batch infer

* Add unit test

* get img_shape as tensor in single_stage

* Fix max_shape

* Update docs and unittest

Co-authored-by: maningsheng <maningsheng@sensetime.com>

* modify get_bboes() to export onnx model which support dynamic input shape

* update code

* update code

* update code

* update code

* update code

* update code

* update code

* update code

* update code

* update code

* make anchor_generator exportable to ONNX

* make k of topk dynamic for onnx

* rename nms_pre_t -> nms_pre_tensor

* bump to v2.10.0

* update other docs

* resolve comments

* resolve comments

* Deprecated warning and fix 'meta' error

* Fix comment

* Deprecated font_scale

* Ignore nonzero and add dummy data in multiclass_nms for deployment with TensorRT

* reduce number of bboxes by nms_pre_deploy in anchor_head

* reduce number of bboxes by nms_pre_deploy in fcos_head

* add onnx2tensorrt and its doc

* add nms_pre_deploy in yolo_head

* change sort to topk in rpn_head

* skip nonzero in single_level_roi_extractor

* update onnx2tensorrt and its doc

* move onnx2tensorrt

* update onnx2tensorrt doc

* change nms_pre_deploy to deploy_nms_pre

* update doc and change deploy_nms_pre format

* add cfg

* add cfg

* fix typo

* fix pad

* fix test pad

* add readme

* modify by comments

* fix typo

* rpn head refactor

* guided anchor

* add deprecate warning

* add more warning for set ori augs

* fix unitest

* change base

* cascade rpn

* cornernet and centri

* grid frcnn ok

* ga anchor ok

* legacy ok

* pisa ok

* tridnet ok

* fix tutorals

* use ConfigDict

* use isinstance method to get loading pipeline

* Fix isinstance error

* Add unit test

* Fix lint

* Fix lint

Co-authored-by: hhaAndroid <1286304229@qq.com>

* add mmcv.jit

* only use derivate and coderize

* fix for isort

* small modify for decorator order

* Add doctest examples to mask structures

* Add doctests examples to mask_target

* Expanded docs and added examples to loss and get_seg_masks in FCNMaskHead

* Add doctest and update docs of mask_cross_entropy

* Modifications based on review comments

* Fixed boxes to be inside the target image

* Fixed doc in cross entropy loss

* moved from after import

* Added tenative scale_factor and rescale docs

* Moved ensure_rng to top level utils

* Added random classmethods to BitmapBoxes and Polygons

* update docstring of get_seg_masks()

* use height / width instead of H, W

* Better scale_factor rescale docs

* fix linting errors to pass CI

Co-authored-by: Kai Chen <chenkaidev@gmail.com>

* sup customize runner

* fix typo

* use build runner

* minor fix

* modify according to comments

* add an assertion

* swap order

* fix tutorial

* avoid crash in empty gt training of GFL

* real fix of the bug

* fix modify args

* add comments

* clean deprecated fp16 api

* clean fp16 import in core

* add bn buffer sync in eval_hook

* fix docstr

* add a comment for modification

* add a abvious import

* add a assert for CLASSES in XMLDataset

* change to get

* add voc tiny dataset

* add tiny voc

* add voc init test

* change to smalle iamge

* add xmlsubclass check

* add smaller image

* add comment for assert

* Support get plt frame data

* Add grayscale image unittest and fix comment

* Clean empty line and fix comment

* Add unit test and fix comment

pls use `iou_thrs` instead of `iou_thr`

* Update inference and add unit test

* Fix lint

* Fix abnormal resource usage

* Update unit test

* feat(api): allow batches in inference API

* fix(code): remove debugging printf

* bump to v2.9.0

* update

* fix num_classes and minor

* update converted models

Code for AAAI 2021 paper. "SCNet: Training Inference Sample Consistency for Instance Segmentation" (#4356)

* add scnet

* add more configs and minors

* fix unittest for scnet

* update docstring

* add forward test for htc and scnet

* fix build on pytorch 1.3 for empty tensor

* update doc string and minor refactor

* update unit-test and use inheritance for SCNet heads

* support scnet tta

* minor

* update readme

* minor fixes

* add CompatibleCheckHook

* check val

* add docstr

* move hook to default runtime

* add unitest

* fix a typo

* fix unitest

* update docs

* add check for CLASSES is None

* fix doc

* Categorize tools

* Update docs

* Rename tools

* Tool organization

* use reduce pos and denorm

* add cfg

* fix cfg

* fix pt1.6

* update link

* clean

* add legacy cfg

* clean cfg

* update readme

* fix zero pos

* update url

* fix: mask resize

* test: mask resize

* chore: fix lint

* draft version train and test

* remove get_bbox in sparse_head

* refactor SparseRCNN model, align inference eval mAP with official repo

* add focal_loss in hungarian_assigner.py

* update hungarian_assigner.py

* update detr config

* update doc and codes

* fix unitest

* fix unitest

* add embedding_rpn and traing code

* modify code format

* fix proposalist in simpletest

* add focal loss traing

* fix acc

* fix reduce mean

* add embedding rpn doc and add show initproposals

* add docstr

* fix backbone lr

* fix typo and change detach_proposal_list to proposal_list

* remove debug assert

* add readme

* add config for random crop when use 300 proposals

* add aug for 300 proposals

* add 300 proposals results and fix docstr

* add 1x results

* fix lint

* fix unitest

* fix docstr of embeddingrpn

* fix according to the comments

* fix get_bbox and a bug of dynamic conv

* support cpu focal loss and fix docstr

* fix docstr

* fix unitest

* fix bug of  init_weight

* fix to device in assign

* fix all commemt

* fix test_config

* remove dim 1 in forward

* fix shape in comment

* fix the case that train_cfg is None

* fix shape of docstr

* fix docstr

* fix typo

* simplify test

* fix docstr

* move train_cfg and test_cfg to model as

* fix unitest

* add link of ckpt and log

* fix 1x results

Co-authored-by: tianyuandu <tianyuandu@gmail.com>
Co-authored-by: fangkairen <fangkairen@sensetime.com>