Function process_batch

evaluation/evaluation_anomaly.py:116–164

Process a batch of images and questions

(model, batch_images, batch_questions, id_list, all_outputs, has_bbox)

        json.dump(all_outputs, f, indent=2, ensure_ascii=False)

def process_batch(model, batch_images, batch_questions, id_list, all_outputs, has_bbox):
    """Process a batch of images and questions"""
    batch_results = model.segment_objects_batch(batch_images, batch_questions)
    
    for i, result in enumerate(batch_results):
        try:
            thinking = result["thinking"]
            bboxes = result["bboxes"]
            mask_all = result["masks"]
            gt_mask = np.array(id_list[i]["mask"])
            
            # visualize_result(batch_images[i], mask_all, gt_mask, thinking, batch_questions[i])
            
            bbox_iou = 0.0
            if has_bbox:
                try:     
                    gt_bbox = id_list[i]["bbox"]
                    for pred_bbox in bboxes:
                        if compute_bbox_iou(pred_bbox, gt_bbox) > 0.5:
                            bbox_iou = 1.0
                            break
                except Exception as e:
                    print(f"Bbox error: {e}, Image ID: {id_list[i]['image_id']}, Ann ID: {id_list[i]['ann_id']}")
                    bbox_iou = 0.0
            
            all_outputs.append({
                "image_id": id_list[i]["image_id"],
                "ann_id": id_list[i]["ann_id"],
                "think": thinking,
                # "mask_pred": mask_all.astype(int).tolist(),
                # "mask_gt": gt_mask.astype(int).tolist(),
                "bbox_iou": bbox_iou,
                "anomaly_pred": float(mask_all.sum() / (id_list[i]["img_height"] * id_list[i]["img_width"])),
                "anomaly_label": int(gt_mask.sum() > 0)
            })
            
        except Exception as e:
            print(f"Error processing result: {e}")
            # Add penalty in this situation
            all_outputs.append({
                "image_id": id_list[i]["image_id"],
                "ann_id": id_list[i]["ann_id"],
                "think": "",
                # "mask_pred": np.zeros_like(id_list[i]["mask"]).astype(int).tolist(),
                # "mask_gt": id_list[i]["mask"].astype(int).tolist(),
                "bbox_iou": 0.0,
                "anomaly_pred": float(id_list[i]["mask"].sum() / (id_list[i]["img_height"] * id_list[i]["img_width"])),
                "anomaly_label": int(id_list[i]["mask"].sum() > 0)
            })

if __name__ == "__main__":
    main()