def _load_annotations()

in vilbert/datasets/guesswhat_pointing_dataset.py [0:0]

• 61 lines of code
• 13 McCabe index (conditional complexity)

    def _load_annotations(self, clean_datasets):
        # Build an index which maps image id with a list of caption annotations.
        entries = []
        remove_ids = []
        if clean_datasets or self.split == "mteval":
            remove_ids = np.load(
                os.path.join(self.dataroot, "cache", "coco_test_ids.npy")
            )
            remove_ids = [int(x) for x in remove_ids]

        all_images = cPickle.load(
            open(os.path.join(self.dataroot, "cache", "image_bbox_list.pkl"), "rb")
        )
        boxes_dict = cPickle.load(
            open(os.path.join(self.dataroot, "cache", "bboxes_dict.pkl"), "rb")
        )

        if self.split == "mteval":
            annotations_path = os.path.join(
                self.dataroot, "guesswhat.%s.jsonl" % "train"
            )
        else:
            annotations_path = os.path.join(
                self.dataroot, "guesswhat.%s.jsonl" % self.split
            )

        with jsonlines.open(annotations_path) as reader:
            # Build an index which maps image id with a list of qa annotations.
            for annotation in reader:
                if (
                    self.split == "train"
                    and int(annotation["image"]["id"]) in remove_ids
                ):
                    continue
                elif (
                    self.split == "mteval"
                    and int(annotation["image"]["id"]) not in remove_ids
                ):
                    continue
                questions = []
                answers = []
                bboxes = []
                for q in annotation["qas"]:
                    questions.append(q["question"])
                    answers.append(q["answer"])

                for o in annotation["objects"]:
                    bboxes.append(o["id"])
                total_bboxes = list(
                    set(all_images[annotation["image"]["id"]]["bboxes"])
                )
                total_bboxes = sorted(total_bboxes)

                bbox_idx = []
                for a in sorted(bboxes):
                    bbox_idx.append(total_bboxes.index(a))

                entries.append(
                    {
                        "questions": questions,
                        "answers": answers,
                        "dialog_id": annotation["id"],
                        "image_id": annotation["image"]["id"],
                        "refBox": boxes_dict[annotation["object_id"]],
                        "ref_id": annotation["object_id"],
                        "mc_idx": bbox_idx,
                    }
                )

        return entries