update

detectron2/config/defaults.py

@@ -675,7 +675,7 @@ _C.MODEL.RESNETS.DEFORM_MODULATED = False
 _C.MODEL.RESNETS.DEFORM_NUM_GROUPS = 1
 
 # ResNest 添加的参数设置
-_C.MODEL.RESNETS.STRIDE_IN_1X1 = False
+# _C.MODEL.RESNETS.STRIDE_IN_1X1 = False
 # Apply deep stem
 _C.MODEL.RESNETS.DEEP_STEM = True
 # Apply avg after conv2 in the BottleBlock

detectron2/data/datasets/builtin.py

@@ -27,7 +27,7 @@ from .cityscapes_panoptic import register_all_cityscapes_panoptic
 from .coco import load_sem_seg, register_coco_instances, register_coco_instances_wsl
 from .coco_panoptic import register_coco_panoptic, register_coco_panoptic_separated
 from .lvis import get_lvis_instances_meta, register_lvis_instances
-from .pascal_voc import register_pascal_voc, register_pascal_voc_wsl, register_pascal_voc_wsl_top1, register_pascal_voc_wsl_thres, register_pascal_voc_wsl_contain, register_pascal_voc_wsl_contain_total, register_pascal_voc_wsl_mist, register_pascal_voc_wsl_mist_contain, register_pascal_voc_wsl_contain_all, register_pascal_voc_wsl_contain_w2f, register_pascal_voc_wsl_oicr_contain, register_pascal_voc_wsl_oicr_contain_all, register_pascal_voc_wsl_w2f_overlap, register_pascal_voc_wsl_contain_all_adaptive
+from .pascal_voc import register_pascal_voc, register_pascal_voc_wsl, register_pascal_voc_wsl_top1, register_pascal_voc_wsl_thres, register_pascal_voc_wsl_contain, register_pascal_voc_wsl_contain_total, register_pascal_voc_wsl_mist, register_pascal_voc_wsl_mist_contain, register_pascal_voc_wsl_contain_all, register_pascal_voc_wsl_contain_w2f, register_pascal_voc_wsl_oicr_contain, register_pascal_voc_wsl_oicr_contain_all, register_pascal_voc_wsl_w2f_overlap, register_pascal_voc_wsl_contain_all_adaptive, register_pascal_voc_wsl_contain_adaptive, register_pascal_voc_wsl_contain_keep, register_pascal_voc_wsl_teach_iter
 
 # ==== Predefined datasets and splits for COCO ==========
 
@@ -133,6 +133,8 @@ COCO_WSL = {
     "coco_2014_valminusminival_casd_wsl_contain": ("coco/val2014", "coco/annotations/casd_valminusminival2014_wsl_contain.json"),
     "coco_2014_train_casd_wsl_contain_adaptive": ("coco/train2014", "coco/annotations/casd_train2014_wsl_contain_adaptive.json"),
     "coco_2014_valminusminival_casd_wsl_contain_adaptive": ("coco/val2014", "coco/annotations/casd_valminusminival2014_wsl_contain_adaptive.json"),
+    "coco_2014_train_casd_wsl_topone": ("coco/train2014", "coco/annotations/casd_train2014_wsl_topone.json"),
+    "coco_2014_valminusminival_casd_wsl_topone": ("coco/val2014", "coco/annotations/casd_valminusminival2014_wsl_topone.json"),
 }
 
 # 注册弱监督的打好 pgt 标记的数据集
@@ -307,6 +309,57 @@ def register_all_pascal_voc_wsl_thres(root):
         register_pascal_voc_wsl_thres(name, os.path.join(root, dirname), split, year)
         MetadataCatalog.get(name).evaluator_type = "pascal_voc"
 
+
+def register_all_pascal_voc_wsl_teach_iter1(root):
+    SPLITS = [
+        ("voc_2007_train_wsl_teach_iter1", "VOC2007", "train"),
+        ("voc_2007_val_wsl_teach_iter1", "VOC2007", "val")
+    ]
+    for name, dirname, split in SPLITS:
+        year = 2007 if "2007" in name else 2012
+        register_pascal_voc_wsl_teach_iter(name, os.path.join(root, dirname), split, year, "1")
+        MetadataCatalog.get(name).evaluator_type = "pascal_voc"
+
+def register_all_pascal_voc_wsl_contain_keep_01(root):
+    SPLITS = [
+        ("voc_2007_train_wsl_contain_keep01", "VOC2007", "train"),
+        ("voc_2007_val_wsl_contain_keep01", "VOC2007", "val")
+    ]
+    for name, dirname, split in SPLITS:
+        year = 2007 if "2007" in name else 2012
+        register_pascal_voc_wsl_contain_keep(name, os.path.join(root, dirname), split, year, "01")
+        MetadataCatalog.get(name).evaluator_type = "pascal_voc"
+
+def register_all_pascal_voc_wsl_contain_keep_015(root):
+    SPLITS = [
+        ("voc_2007_train_wsl_contain_keep015", "VOC2007", "train"),
+        ("voc_2007_val_wsl_contain_keep015", "VOC2007", "val")
+    ]
+    for name, dirname, split in SPLITS:
+        year = 2007 if "2007" in name else 2012
+        register_pascal_voc_wsl_contain_keep(name, os.path.join(root, dirname), split, year, "015")
+        MetadataCatalog.get(name).evaluator_type = "pascal_voc"
+
+def register_all_pascal_voc_wsl_contain_keep_025(root):
+    SPLITS = [
+        ("voc_2007_train_wsl_contain_keep025", "VOC2007", "train"),
+        ("voc_2007_val_wsl_contain_keep025", "VOC2007", "val")
+    ]
+    for name, dirname, split in SPLITS:
+        year = 2007 if "2007" in name else 2012
+        register_pascal_voc_wsl_contain_keep(name, os.path.join(root, dirname), split, year, "025")
+        MetadataCatalog.get(name).evaluator_type = "pascal_voc"
+
+def register_all_pascal_voc_wsl_contain_keep_03(root):
+    SPLITS = [
+        ("voc_2007_train_wsl_contain_keep03", "VOC2007", "train"),
+        ("voc_2007_val_wsl_contain_keep03", "VOC2007", "val")
+    ]
+    for name, dirname, split in SPLITS:
+        year = 2007 if "2007" in name else 2012
+        register_pascal_voc_wsl_contain_keep(name, os.path.join(root, dirname), split, year, "03")
+        MetadataCatalog.get(name).evaluator_type = "pascal_voc"
+
 def register_all_pascal_voc_wsl_contain_075(root):
     SPLITS = [
         ("voc_2007_train_wsl_contain_075", "VOC2007", "train"),
@@ -417,6 +470,16 @@ def register_all_pascal_voc_wsl_contain_all_adaptive(root):
         register_pascal_voc_wsl_contain_all_adaptive(name, os.path.join(root, dirname), split, year)
         MetadataCatalog.get(name).evaluator_type = "pascal_voc"
 
+def register_all_pascal_voc_wsl_contain_adaptive(root):
+    SPLITS = [
+        ("voc_2007_train_wsl_contain_adaptive", "VOC2007", "train"),
+        ("voc_2007_val_wsl_contain_adaptive", "VOC2007", "val")
+    ]
+    for name, dirname, split in SPLITS:
+        year = 2007 if "2007" in name else 2012
+        register_pascal_voc_wsl_contain_adaptive(name, os.path.join(root, dirname), split, year)
+        MetadataCatalog.get(name).evaluator_type = "pascal_voc"
+
 def register_all_pascal_voc_w2f(root):
     SPLITS = [
         ("voc_2007_train_wsl_w2f", "VOC2007", "train"),
@@ -482,4 +545,10 @@ if __name__.endswith(".builtin"):
     register_all_pascal_voc_wsl_oicr_contain(_root)
     register_all_pascal_voc_wsl_oicr_contain_all(_root)
     register_all_pascal_voc_wsl_contain_all_adaptive(_root)
+    register_all_pascal_voc_wsl_contain_adaptive(_root)
+    register_all_pascal_voc_wsl_contain_keep_01(_root)
+    register_all_pascal_voc_wsl_contain_keep_015(_root)
+    register_all_pascal_voc_wsl_contain_keep_025(_root)
+    register_all_pascal_voc_wsl_contain_keep_03(_root)
+    register_all_pascal_voc_wsl_teach_iter1(_root)
     register_all_ade20k(_root)

detectron2/data/datasets/pascal_voc.py

@@ -12,7 +12,7 @@ from detectron2.data import DatasetCatalog, MetadataCatalog
 from detectron2.structures import BoxMode
 from detectron2.utils.file_io import PathManager
 
-__all__ = ["load_voc_instances", "register_pascal_voc", "register_pascal_voc_wsl", "register_pascal_voc_wsl_top1", "register_pascal_voc_wsl_thres", "register_pascal_voc_wsl_mist", "register_pascal_voc_wsl_mist_contain", "register_pascal_voc_wsl_contain_all", "register_pascal_voc_wsl_contain_w2f", "register_pascal_voc_wsl_oicr_contain", "register_pascal_voc_wsl_oicr_contain_all", "register_pascal_voc_wsl_w2f_overlap", "register_pascal_voc_wsl_contain_all_adaptive"]
+__all__ = ["load_voc_instances", "register_pascal_voc", "register_pascal_voc_wsl", "register_pascal_voc_wsl_top1", "register_pascal_voc_wsl_thres", "register_pascal_voc_wsl_mist", "register_pascal_voc_wsl_mist_contain", "register_pascal_voc_wsl_contain_all", "register_pascal_voc_wsl_contain_w2f", "register_pascal_voc_wsl_oicr_contain", "register_pascal_voc_wsl_oicr_contain_all", "register_pascal_voc_wsl_w2f_overlap", "register_pascal_voc_wsl_contain_all_adaptive", "register_pascal_voc_wsl_contain_adaptive", "register_pascal_voc_wsl_contain_keep", "register_pascal_voc_wsl_teach_iter"]
 
 
 # fmt: off
@@ -33,6 +33,7 @@ def load_voc_instances(dirname: str, split: str, class_names: Union[List[str], T
         split (str): one of "train", "test", "val", "trainval"
         class_names: list or tuple of class names
     """
+    # print(1)
     with PathManager.open(os.path.join(dirname, "ImageSets", "Main", split + ".txt")) as f:
         fileids = np.loadtxt(f, dtype=np.str)
 
@@ -89,7 +90,7 @@ def load_voc_instances_wsl(dirname: str, split: str, class_names: Union[List[str
     # 获取 数据集对应划分(train, val, test) 图片 ids
     with PathManager.open(os.path.join(dirname, "ImageSets", "Main", split + ".txt")) as f:
         fileids = np.loadtxt(f, dtype=np.str)
-    
+    # print(2)
     # 针对 single-input 的文件
     # print("load from {}/single_voc07_wsl_{}_contain.json".format(dirname, split))
     # annotation_wsl = json.load(open(
@@ -470,6 +471,163 @@ def load_voc_instances_wsl_contain(dirname: str, split: str, thres, class_names:
         dicts.append(r)
     return dicts
 
+def load_voc_instances_wsl_contain_keep(dirname: str, split: str, thres, class_names: Union[List[str], Tuple[str, ...]]):
+    # 获取 数据集对应划分(train, val, test) 图片 ids
+    with PathManager.open(os.path.join(dirname, "ImageSets", "Main", split + ".txt")) as f:
+        fileids = np.loadtxt(f, dtype=np.str)
+    
+    # 针对 single-input 的文件
+    # print("load from {}/single_voc07_wsl_{}_contain.json".format(dirname, split))
+    # annotation_wsl = json.load(open(
+    #    "{}/single_voc07_wsl_{}_contain.json".format(dirname, split), "r"
+    # ))
+    # 获取 annotations, wsl 预测之后的结果会保存为 json 的格式
+    if "07" in dirname:
+        annotation_wsl = json.load(open(
+            "{}/voc07_wsl_{}_contain_keep{}.json".format(dirname, split, thres), "r"
+        ))
+    elif "12" in dirname:
+        annotation_wsl = json.load(open(
+            "{}/casd_voc12_wsl_{}_contain_keep{}.json".format(dirname, split, thres), "r"
+        ))
+    else:
+        assert False, "Wrong dirname: {}".format(dirname)
+    multi_class_labels = None
+    if "multi_label" in annotation_wsl:
+        multi_class_labels = annotation_wsl.pop("multi_label")
+    annotation_dirname = PathManager.get_local_path(os.path.join(dirname, "Annotations/"))
+    dicts = []
+    for fileid in fileids:
+        anno = annotation_wsl[str(int(fileid))]
+        jpeg_file = os.path.join(dirname, "JPEGImages", fileid + ".jpg")
+        anno_file = os.path.join(annotation_dirname, fileid + ".xml")
+
+        if not os.path.isfile(anno_file):
+            with Image.open(jpeg_file) as img:
+                width, height = img.size
+            r = {"file_name": jpeg_file, "image_id": fileid, "height": height, "width": width}
+            instances = []
+            for obj in anno:
+                bbox = obj["bbox"]
+                bbox = [int(i) for i in bbox] # 因为 predict 出来的 bbox 是float, 要转化为 int list
+                category_id = obj["category_id"] - 1 # 因为保存统计时将 index + 1 了从而方便 TIDE 统计了, 因此这里需要 - 1
+                instances.append(
+                    {
+                        "category_id": category_id, "bbox": bbox, "bbox_mode": BoxMode.XYXY_ABS
+                    }
+                )
+            r["annotations"] = instances
+            if multi_class_labels is not None:
+                r["multi_label"] = multi_class_labels[str(int(fileid))]
+            dicts.append(r)
+            continue
+
+        with PathManager.open(anno_file) as f:
+            tree = ET.parse(f)
+
+        r = {
+            "file_name": jpeg_file,
+            "image_id": fileid,
+            "height": int(tree.findall("./size/height")[0].text),
+            "width": int(tree.findall("./size/width")[0].text),
+        }
+        instances = []
+        # 这里是从 annotation_wsl 中进行 gt 信息的提取, 而不是 从 anno file 中提取真正的 gt 信息出来
+        for obj in anno:
+            bbox = obj["bbox"]
+            bbox = [int(i) for i in bbox]
+            category_id = obj["category_id"] - 1
+            instances.append(
+                {
+                    "category_id": category_id, "bbox": bbox, "bbox_mode": BoxMode.XYXY_ABS
+                }
+            )
+        r["annotations"] = instances
+        if multi_class_labels is not None:
+            r["multi_label"] = multi_class_labels[str(int(fileid))]
+        dicts.append(r)
+    return dicts
+
+
+def load_voc_instances_wsl_teach_iter(dirname: str, split: str, thres, class_names: Union[List[str], Tuple[str, ...]]):
+    # 获取 数据集对应划分(train, val, test) 图片 ids
+    with PathManager.open(os.path.join(dirname, "ImageSets", "Main", split + ".txt")) as f:
+        fileids = np.loadtxt(f, dtype=np.str)
+    
+    # 针对 single-input 的文件
+    # print("load from {}/single_voc07_wsl_{}_contain.json".format(dirname, split))
+    # annotation_wsl = json.load(open(
+    #    "{}/single_voc07_wsl_{}_contain.json".format(dirname, split), "r"
+    # ))
+    # 获取 annotations, wsl 预测之后的结果会保存为 json 的格式
+    if "07" in dirname:
+        annotation_wsl = json.load(open(
+            "{}/voc07_wsl_{}_teach_iter{}.json".format(dirname, split, thres), "r"
+        ))
+    elif "12" in dirname:
+        annotation_wsl = json.load(open(
+            "{}/casd_voc12_wsl_{}_teach_iter{}.json".format(dirname, split, thres), "r"
+        ))
+    else:
+        assert False, "Wrong dirname: {}".format(dirname)
+    multi_class_labels = None
+    if "multi_label" in annotation_wsl:
+        multi_class_labels = annotation_wsl.pop("multi_label")
+    annotation_dirname = PathManager.get_local_path(os.path.join(dirname, "Annotations/"))
+    dicts = []
+    for fileid in fileids:
+        if str(int(fileid)) not in annotation_wsl:
+            continue
+        anno = annotation_wsl[str(int(fileid))]
+        jpeg_file = os.path.join(dirname, "JPEGImages", fileid + ".jpg")
+        anno_file = os.path.join(annotation_dirname, fileid + ".xml")
+
+        if not os.path.isfile(anno_file):
+            with Image.open(jpeg_file) as img:
+                width, height = img.size
+            r = {"file_name": jpeg_file, "image_id": fileid, "height": height, "width": width}
+            instances = []
+            for obj in anno:
+                bbox = obj["bbox"]
+                bbox = [int(i) for i in bbox] # 因为 predict 出来的 bbox 是float, 要转化为 int list
+                category_id = obj["category_id"] - 1 # 因为保存统计时将 index + 1 了从而方便 TIDE 统计了, 因此这里需要 - 1
+                instances.append(
+                    {
+                        "category_id": category_id, "bbox": bbox, "bbox_mode": BoxMode.XYXY_ABS
+                    }
+                )
+            r["annotations"] = instances
+            if multi_class_labels is not None:
+                r["multi_label"] = multi_class_labels[str(int(fileid))]
+            dicts.append(r)
+            continue
+
+        with PathManager.open(anno_file) as f:
+            tree = ET.parse(f)
+
+        r = {
+            "file_name": jpeg_file,
+            "image_id": fileid,
+            "height": int(tree.findall("./size/height")[0].text),
+            "width": int(tree.findall("./size/width")[0].text),
+        }
+        instances = []
+        # 这里是从 annotation_wsl 中进行 gt 信息的提取, 而不是 从 anno file 中提取真正的 gt 信息出来
+        for obj in anno:
+            bbox = obj["bbox"]
+            bbox = [int(i) for i in bbox]
+            category_id = obj["category_id"] - 1
+            instances.append(
+                {
+                    "category_id": category_id, "bbox": bbox, "bbox_mode": BoxMode.XYXY_ABS
+                }
+            )
+        r["annotations"] = instances
+        if multi_class_labels is not None:
+            r["multi_label"] = multi_class_labels[str(int(fileid))]
+        dicts.append(r)
+    return dicts
+
 def load_voc_instances_wsl_contain_all(dirname: str, split: str, class_names: Union[List[str], Tuple[str, ...]]):
     # 获取 数据集对应划分(train, val, test) 图片 ids
     with PathManager.open(os.path.join(dirname, "ImageSets", "Main", split + ".txt")) as f:
@@ -626,6 +784,85 @@ def load_voc_instances_wsl_contain_all_adaptive(dirname: str, split: str, class_
         dicts.append(r)
     return dicts
 
+def load_voc_instances_wsl_contain_adaptive(dirname: str, split: str, class_names: Union[List[str], Tuple[str, ...]]):
+    # 获取 数据集对应划分(train, val, test) 图片 ids
+    with PathManager.open(os.path.join(dirname, "ImageSets", "Main", split + ".txt")) as f:
+        fileids = np.loadtxt(f, dtype=np.str)
+    
+    # 针对 single-input 的文件
+    # print("load from {}/single_voc07_wsl_{}_contain.json".format(dirname, split))
+    # annotation_wsl = json.load(open(
+    #    "{}/single_voc07_wsl_{}_contain.json".format(dirname, split), "r"
+    # ))
+    # 获取 annotations, wsl 预测之后的结果会保存为 json 的格式
+    if "07" in dirname:
+        annotation_wsl = json.load(open(
+            "{}/voc07_wsl_{}_contain_adaptive.json".format(dirname, split), "r"
+        ))
+    elif "12" in dirname:
+        annotation_wsl = json.load(open(
+            "{}/casd_voc12_wsl_{}_contain_adaptive.json".format(dirname, split), "r"
+        ))
+    else:
+        assert False, "Wrong dirname: {}".format(dirname)
+    multi_class_labels = None
+    if "multi_label" in annotation_wsl:
+        multi_class_labels = annotation_wsl.pop("multi_label")
+    annotation_dirname = PathManager.get_local_path(os.path.join(dirname, "Annotations/"))
+    dicts = []
+    for fileid in fileids:
+        if str(int(fileid)) not in annotation_wsl:
+            continue
+        anno = annotation_wsl[str(int(fileid))]
+        jpeg_file = os.path.join(dirname, "JPEGImages", fileid + ".jpg")
+        anno_file = os.path.join(annotation_dirname, fileid + ".xml")
+
+        if not os.path.isfile(anno_file):
+            with Image.open(jpeg_file) as img:
+                width, height = img.size
+            r = {"file_name": jpeg_file, "image_id": fileid, "height": height, "width": width}
+            instances = []
+            for obj in anno:
+                bbox = obj["bbox"]
+                bbox = [int(i) for i in bbox] # 因为 predict 出来的 bbox 是float, 要转化为 int list
+                category_id = obj["category_id"] - 1 # 因为保存统计时将 index + 1 了从而方便 TIDE 统计了, 因此这里需要 - 1
+                instances.append(
+                    {
+                        "category_id": category_id, "bbox": bbox, "bbox_mode": BoxMode.XYXY_ABS
+                    }
+                )
+            r["annotations"] = instances
+            if multi_class_labels is not None:
+                r["multi_label"] = multi_class_labels[str(int(fileid))]
+            dicts.append(r)
+            continue
+
+        with PathManager.open(anno_file) as f:
+            tree = ET.parse(f)
+
+        r = {
+            "file_name": jpeg_file,
+            "image_id": fileid,
+            "height": int(tree.findall("./size/height")[0].text),
+            "width": int(tree.findall("./size/width")[0].text),
+        }
+        instances = []
+        # 这里是从 annotation_wsl 中进行 gt 信息的提取, 而不是 从 anno file 中提取真正的 gt 信息出来
+        for obj in anno:
+            bbox = obj["bbox"]
+            bbox = [int(i) for i in bbox]
+            category_id = obj["category_id"] - 1
+            instances.append(
+                {
+                    "category_id": category_id, "bbox": bbox, "bbox_mode": BoxMode.XYXY_ABS
+                }
+            )
+        r["annotations"] = instances
+        if multi_class_labels is not None:
+            r["multi_label"] = multi_class_labels[str(int(fileid))]
+        dicts.append(r)
+    return dicts
+
 def load_voc_instances_wsl_w2f(dirname: str, split: str, class_names: Union[List[str], Tuple[str, ...]]):
     # 获取 数据集对应划分(train, val, test) 图片 ids
     with PathManager.open(os.path.join(dirname, "ImageSets", "Main", split + ".txt")) as f:
@@ -1069,6 +1306,18 @@ def register_pascal_voc_wsl_contain(name, dirname, split, year, thres, class_nam
         thing_classes=list(class_names), dirname=dirname, year=year, split=split
     )
 
+def register_pascal_voc_wsl_contain_keep(name, dirname, split, year, thres, class_names=CLASS_NAMES):
+    DatasetCatalog.register(name, lambda: load_voc_instances_wsl_contain_keep(dirname, split, thres, class_names))
+    MetadataCatalog.get(name).set(
+        thing_classes=list(class_names), dirname=dirname, year=year, split=split
+    )
+
+def register_pascal_voc_wsl_teach_iter(name, dirname, split, year, thres, class_names=CLASS_NAMES):
+    DatasetCatalog.register(name, lambda: load_voc_instances_wsl_teach_iter(dirname, split, thres, class_names))
+    MetadataCatalog.get(name).set(
+        thing_classes=list(class_names), dirname=dirname, year=year, split=split
+    )
+
 def register_pascal_voc_wsl_contain_total(name, dirname, split, year, class_names=CLASS_NAMES):
     DatasetCatalog.register(name, lambda: load_voc_instances_wsl_contain_total(dirname, split, class_names))
     MetadataCatalog.get(name).set(
@@ -1112,6 +1361,12 @@ def register_pascal_voc_wsl_contain_all_adaptive(name, dirname, split, year, cla
         thing_classes=list(class_names), dirname=dirname, year=year, split=split
     )
 
+def register_pascal_voc_wsl_contain_adaptive(name, dirname, split, year, class_names=CLASS_NAMES):
+    DatasetCatalog.register(name, lambda: load_voc_instances_wsl_contain_adaptive(dirname, split, class_names))
+    MetadataCatalog.get(name).set(
+        thing_classes=list(class_names), dirname=dirname, year=year, split=split
+    )
+
 def register_pascal_voc_wsl_contain_w2f(name, dirname, split, year, class_names=CLASS_NAMES):
     DatasetCatalog.register(name, lambda: load_voc_instances_wsl_w2f(dirname, split, class_names))
     MetadataCatalog.get(name).set(

detectron2/evaluation/pascal_voc_evaluation.py

@@ -29,6 +29,7 @@ class PascalVOCDetectionEvaluator(DatasetEvaluator):
     """
 
     def __init__(self, dataset_name):
+        # PS: 之所以会出现使用不同数据集结果 test 结果相同的原因好像是这里会根据 annotation 文件重新读一遍 gt 信息, 而不是用 dataset 中记录的
         """
         Args:
             dataset_name (str): name of the dataset, e.g., "voc_2007_test"
@@ -77,25 +78,34 @@ class PascalVOCDetectionEvaluator(DatasetEvaluator):
         if not comm.is_main_process():
             return
         # 只有单卡测试的时候将检测结果进行记录
-        # import json
+        import json
         # # print(all_predictions)
-        # if len(all_predictions) == 1:
-        #     all_prediction = all_predictions[0]
-        #     print("???")
-        #     tide_result = []
-        #     for cls_id, class_name in enumerate(self._class_names):
-        #         lines = all_prediction[cls_id]
-        #         for line in lines:
-        #             message = line.split(" ")
-        #             tide_result.append(
-        #                 {
-        #                     "image_id": int(message[0]),
-        #                     "category_id": cls_id+1,
-        #                     "score": float(message[1]),
-        #                     "bbox": [float(message[2]), float(message[3]), float(message[4]), float(message[5])]
-        #                 }
-        #             )
-        #     json.dump(tide_result, open("/mnt/data3/suilin/wsod/visual/{}.json".format(self._dataset_name), "w"))
+        if len(all_predictions) == 1:
+            all_prediction = all_predictions[0]
+        else:
+            all_prediction = {}
+            for p in all_predictions:
+                for key in list(p.keys()):
+                    if key not in all_prediction:
+                        all_prediction[key] = p[key]
+                    else:
+                        all_prediction[key] = all_prediction[key] + p[key]
+
+        tide_result = []
+        for cls_id, class_name in enumerate(self._class_names):
+            lines = all_prediction[cls_id]
+            for line in lines:
+                message = line.split(" ")
+                tide_result.append(
+                    {
+                        "image_id": int(message[0]),
+                        "category_id": cls_id+1,
+                        "score": float(message[1]),
+                        "bbox": [float(message[2]), float(message[3]), float(message[4]), float(message[5])]
+                    }
+                )
+        # json.dump(tide_result, open("/mnt/data3/suilin/wsod/visual/{}.json".format(self._dataset_name), "w"))
+        # json.dump(tide_result, open("/home/suil/codes/github/vis/{}_teach_iter1.json".format(self._dataset_name), "w"))
 
         # else:
         #     pass
@@ -478,4 +488,4 @@ def voc_eval_corloc(detpath, annopath, imagesetfile, classname, ovthresh=0.5, us
         else:
             F.append(image_ids[d])
 
-    return 1.0 * len(T) / npos_im
\ No newline at end of file
+    return 1.0 * len(T) / npos_im