Function process_batch

evaluation/evaluation_segmentation.py:157–223

Process a batch of images and questions

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

    return [int(x_min), int(y_min), int(x_max), int(y_max)]

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])
            
            intersection, union = compute_iou(mask_all, gt_mask)
            
            bbox_iou = 0.0
            
            try:     
                if has_bbox:
                    gt_bbox = id_list[i]["bbox"]
                else:
                    gt_bbox = get_bbox(gt_mask)
                
                if gt_bbox == []:
                    if len(bboxes) == 0:
                        bbox_iou = 1.0
                    else:
                        bbox_iou = 0.0
                else:
                    # adapt to multi-object detection
                    for pred_bbox in bboxes:
                        if compute_bbox_iou(pred_bbox, gt_bbox) > 0.5:
                            bbox_iou = 1.0
                            break
                    merged_bbox = merge_bboxes(bboxes)
                    if merged_bbox:
                        if compute_bbox_iou(merged_bbox, gt_bbox) > 0.5:
                            bbox_iou = 1.0
                    
            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,
                "intersection": int(intersection),
                "union": int(union),
                "bbox_iou": bbox_iou,
                "non_object": int(gt_mask.sum()==0 and len(bboxes)==0),   # for grefcoco where there is no object in the image
                "non_object_GT": int(gt_mask.sum()==0)
            })
            
        except Exception as e:
            print(f"Error processing result: {e}")
            # Add penalty in this situation
            all_outputs.append({