WZMIAOMIAO/deep-learning-for-image-processing · error · KeyError

not support iou_type: {self.iou_type}

Error message

not support iou_type: {self.iou_type}

What it means

CocoEvaluator.update only supports the iou_types 'bbox' and 'segm'; anything else (e.g. 'keypoints') has no prepare method wired up and raises KeyError. Note the message says 'not support' despite being a KeyError.

Source

Thrown at pytorch_object_detection/mask_rcnn/train_utils/coco_eval.py:130

                class_idx = int(label)
                if self.classes_mapping is not None:
                    class_idx = int(self.classes_mapping[str(class_idx)])

                res = {"image_id": img_id,
                       "category_id": class_idx,
                       "segmentation": rle,
                       "score": round(score, 3)}
                res_list.append(res)
            self.results.append(res_list)

    def update(self, targets, outputs):
        if self.iou_type == "bbox":
            self.prepare_for_coco_detection(targets, outputs)
        elif self.iou_type == "segm":
            self.prepare_for_coco_segmentation(targets, outputs)
        else:
            raise KeyError(f"not support iou_type: {self.iou_type}")

    def synchronize_results(self):
        # 同步所有进程中的数据
        eval_ids, eval_results = merge(self.img_ids, self.results)
        self.aggregation_results = {"img_ids": eval_ids, "results": eval_results}

        # 主进程上保存即可
        if is_main_process():
            results = []
            [results.extend(i) for i in eval_results]
            # write predict results into json file
            json_str = json.dumps(results, indent=4)
            with open(self.results_file_name, 'w') as json_file:
                json_file.write(json_str)

    def evaluate(self):
        # 只在主进程上评估即可
        if is_main_process():

View on GitHub (pinned to 1ec3fe6f37)

Solutions

1. Restrict iou_types to 'bbox' or 'segm' when building CocoEvaluator
2. Fix the typo in the iou_type string
3. Add a prepare_for_coco_keypoints-style branch in coco_eval.py if you need keypoints

Example fix

// before
evaluator = CocoEvaluator(coco_gt, iou_types=["keypoints"])
// after
evaluator = CocoEvaluator(coco_gt, iou_types=["bbox", "segm"])

SUPPORTED = {'bbox', 'segm'}
assert set(iou_types).issubset(SUPPORTED), f"iou_types must be subset of {SUPPORTED}"
evaluator = CocoEvaluator(coco_gt, iou_types=iou_types)

def supported_iou_types(types):
    return all(t in ('bbox', 'segm') for t in types)

try:
    evaluator.update(targets, outputs)
except KeyError as e:
    if 'iou_type' in str(e): logger.error('unsupported iou_type; use bbox or segm')
    raise

When it happens

Trigger: Constructing CocoEvaluator(coco_gt, iou_types=["keypoints"]) or any unsupported type, then calling update(targets, outputs).

Common situations: Copying detection code to a pose-estimation task; typo in iou_type like 'bboxs' or 'segmentation'; passing a list element from torchvision defaults that this vendored copy doesn't implement.

Related errors

• Unknown iou type {}
• Please run accumulate() first
• not support iou_type: {self.iou_type}
• Please run accumulate() first
• Please run accumulate() first

AI-assisted analysis of WZMIAOMIAO/deep-learning-for-image-processing@1ec3fe6f37 (2026-08-30). Data as JSON: /api/errors/6ab5595ce09bde8d. Report an issue: GitHub.