def _process_features()

in tools/scripts/features/frcnn/extract_features_frcnn.py [0:0]

• 87 lines of code
• 6 McCabe index (conditional complexity)

    def _process_features(self, features, index):
        feature_keys = [
            "obj_ids",
            "obj_probs",
            "attr_ids",
            "attr_probs",
            "boxes",
            "sizes",
            "preds_per_image",
            "roi_features",
            "normalized_boxes",
        ]
        single_features = dict()

        for key in feature_keys:
            single_features[key] = features[key][index]

        confidence = self.args.confidence_threshold
        idx = 0
        while idx < single_features["obj_ids"].size()[0]:
            removed = False
            if (
                single_features["obj_probs"][idx] < confidence
                or single_features["attr_probs"][idx] < confidence
            ):
                single_features["obj_ids"] = torch.cat(
                    [
                        single_features["obj_ids"][0:idx],
                        single_features["obj_ids"][idx + 1 :],
                    ]
                )
                single_features["obj_probs"] = torch.cat(
                    [
                        single_features["obj_probs"][0:idx],
                        single_features["obj_probs"][idx + 1 :],
                    ]
                )
                single_features["attr_ids"] = torch.cat(
                    [
                        single_features["attr_ids"][0:idx],
                        single_features["attr_ids"][idx + 1 :],
                    ]
                )
                single_features["attr_probs"] = torch.cat(
                    [
                        single_features["attr_probs"][0:idx],
                        single_features["attr_probs"][idx + 1 :],
                    ]
                )
                single_features["boxes"] = torch.cat(
                    [
                        single_features["boxes"][0:idx, :],
                        single_features["boxes"][idx + 1 :, :],
                    ]
                )
                single_features["preds_per_image"] = (
                    single_features["preds_per_image"] - 1
                )
                single_features["roi_features"] = torch.cat(
                    [
                        single_features["roi_features"][0:idx, :],
                        single_features["roi_features"][idx + 1 :, :],
                    ]
                )
                single_features["normalized_boxes"] = torch.cat(
                    [
                        single_features["normalized_boxes"][0:idx, :],
                        single_features["normalized_boxes"][idx + 1 :, :],
                    ]
                )
                removed = True
            if not removed:
                idx += 1

        feat_list = single_features["roi_features"]

        boxes = single_features["boxes"][: self.args.num_features].cpu().numpy()
        num_boxes = self.args.num_features
        objects = single_features["obj_ids"][: self.args.num_features].cpu().numpy()
        probs = single_features["obj_probs"][: self.args.num_features].cpu().numpy()
        width = single_features["sizes"][1].item()
        height = single_features["sizes"][0].item()
        info_list = {
            "bbox": boxes,
            "num_boxes": num_boxes,
            "objects": objects,
            "cls_prob": probs,
            "image_width": width,
            "image_height": height,
        }

        return single_features, feat_list, info_list