2025-07-07 11:04:05,796 - PropVG - INFO - dataset = 'RefCOCOUNC' data_root = './data/seqtr_type/' img_norm_cfg = dict( mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375]) train_pipeline = [ dict( type='LoadImageAnnotationsFromFile_TO', max_token=20, with_mask=True, with_bbox=True, dataset='RefCOCOUNC', use_token_type='beit3', refer_file='data/seqtr_type/annotations/mixed-seg/coco_all.json', object_area_filter=100, object_area_rate_filter=[0.05, 0.8]), dict(type='Resize', img_scale=(384, 384), keep_ratio=False), dict( type='Normalize', mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375]), dict(type='DefaultFormatBundle'), dict( type='CollectData', keys=[ 'img', 'ref_expr_inds', 'text_attention_mask', 'gt_mask_rle', 'gt_bbox' ], meta_keys=[ 'filename', 'expression', 'ori_shape', 'img_shape', 'pad_shape', 'scale_factor', 'gt_ori_mask', 'target', 'empty', 'refer_target_index' ]) ] val_pipeline = [ dict( type='LoadImageAnnotationsFromFile_TO', max_token=20, with_mask=True, with_bbox=True, dataset='RefCOCOUNC', use_token_type='beit3', refer_file='data/seqtr_type/annotations/mixed-seg/coco_all.json', object_area_filter=100, object_area_rate_filter=[0.05, 0.8]), dict(type='Resize', img_scale=(384, 384), keep_ratio=False), dict( type='Normalize', mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375]), dict(type='DefaultFormatBundle'), dict( type='CollectData', keys=[ 'img', 'ref_expr_inds', 'text_attention_mask', 'gt_mask_rle', 'gt_bbox' ], meta_keys=[ 'filename', 'expression', 'ori_shape', 'img_shape', 'pad_shape', 'scale_factor', 'gt_ori_mask', 'target', 'empty', 'refer_target_index' ]) ] test_pipeline = [ dict( type='LoadImageAnnotationsFromFile_TO', max_token=20, with_mask=True, with_bbox=True, dataset='RefCOCOUNC', use_token_type='beit3', refer_file='data/seqtr_type/annotations/mixed-seg/coco_all.json', object_area_filter=100, object_area_rate_filter=[0.05, 0.8]), dict(type='Resize', img_scale=(384, 384), keep_ratio=False), dict( type='Normalize', mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375]), dict(type='DefaultFormatBundle'), dict( type='CollectData', keys=[ 'img', 'ref_expr_inds', 'text_attention_mask', 'gt_mask_rle', 'gt_bbox' ], meta_keys=[ 'filename', 'expression', 'ori_shape', 'img_shape', 'pad_shape', 'scale_factor', 'gt_ori_mask', 'target', 'empty', 'refer_target_index' ]) ] word_emb_cfg = dict(type='GloVe') data = dict( samples_per_gpu=8, workers_per_gpu=4, train=dict( type='RefCOCOUNC', which_set='train', img_source=['coco'], annsfile= './data/seqtr_type/annotations/refcoco-unc/instances_withid.json', imgsfile='./data/seqtr_type/images/mscoco/train2014', pipeline=[ dict( type='LoadImageAnnotationsFromFile_TO', max_token=20, with_mask=True, with_bbox=True, dataset='RefCOCOUNC', use_token_type='beit3', refer_file= 'data/seqtr_type/annotations/mixed-seg/coco_all.json', object_area_filter=100, object_area_rate_filter=[0.05, 0.8]), dict(type='Resize', img_scale=(384, 384), keep_ratio=False), dict( type='Normalize', mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375]), dict(type='DefaultFormatBundle'), dict( type='CollectData', keys=[ 'img', 'ref_expr_inds', 'text_attention_mask', 'gt_mask_rle', 'gt_bbox' ], meta_keys=[ 'filename', 'expression', 'ori_shape', 'img_shape', 'pad_shape', 'scale_factor', 'gt_ori_mask', 'target', 'empty', 'refer_target_index' ]) ], word_emb_cfg=dict(type='GloVe')), val=dict( type='RefCOCOUNC', which_set='val', img_source=['coco'], annsfile= './data/seqtr_type/annotations/refcoco-unc/instances_withid.json', imgsfile='./data/seqtr_type/images/mscoco/train2014', pipeline=[ dict( type='LoadImageAnnotationsFromFile_TO', max_token=20, with_mask=True, with_bbox=True, dataset='RefCOCOUNC', use_token_type='beit3', refer_file= 'data/seqtr_type/annotations/mixed-seg/coco_all.json', object_area_filter=100, object_area_rate_filter=[0.05, 0.8]), dict(type='Resize', img_scale=(384, 384), keep_ratio=False), dict( type='Normalize', mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375]), dict(type='DefaultFormatBundle'), dict( type='CollectData', keys=[ 'img', 'ref_expr_inds', 'text_attention_mask', 'gt_mask_rle', 'gt_bbox' ], meta_keys=[ 'filename', 'expression', 'ori_shape', 'img_shape', 'pad_shape', 'scale_factor', 'gt_ori_mask', 'target', 'empty', 'refer_target_index' ]) ], word_emb_cfg=dict(type='GloVe')), testA=dict( type='RefCOCOUNC', which_set='testA', img_source=['coco'], annsfile= './data/seqtr_type/annotations/refcoco-unc/instances_withid.json', imgsfile='./data/seqtr_type/images/mscoco/train2014', pipeline=[ dict( type='LoadImageAnnotationsFromFile_TO', max_token=20, with_mask=True, with_bbox=True, dataset='RefCOCOUNC', use_token_type='beit3', refer_file= 'data/seqtr_type/annotations/mixed-seg/coco_all.json', object_area_filter=100, object_area_rate_filter=[0.05, 0.8]), dict(type='Resize', img_scale=(384, 384), keep_ratio=False), dict( type='Normalize', mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375]), dict(type='DefaultFormatBundle'), dict( type='CollectData', keys=[ 'img', 'ref_expr_inds', 'text_attention_mask', 'gt_mask_rle', 'gt_bbox' ], meta_keys=[ 'filename', 'expression', 'ori_shape', 'img_shape', 'pad_shape', 'scale_factor', 'gt_ori_mask', 'target', 'empty', 'refer_target_index' ]) ], word_emb_cfg=dict(type='GloVe')), testB=dict( type='RefCOCOUNC', which_set='testB', img_source=['coco'], annsfile= './data/seqtr_type/annotations/refcoco-unc/instances_withid.json', imgsfile='./data/seqtr_type/images/mscoco/train2014', pipeline=[ dict( type='LoadImageAnnotationsFromFile_TO', max_token=20, with_mask=True, with_bbox=True, dataset='RefCOCOUNC', use_token_type='beit3', refer_file= 'data/seqtr_type/annotations/mixed-seg/coco_all.json', object_area_filter=100, object_area_rate_filter=[0.05, 0.8]), dict(type='Resize', img_scale=(384, 384), keep_ratio=False), dict( type='Normalize', mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375]), dict(type='DefaultFormatBundle'), dict( type='CollectData', keys=[ 'img', 'ref_expr_inds', 'text_attention_mask', 'gt_mask_rle', 'gt_bbox' ], meta_keys=[ 'filename', 'expression', 'ori_shape', 'img_shape', 'pad_shape', 'scale_factor', 'gt_ori_mask', 'target', 'empty', 'refer_target_index' ]) ], word_emb_cfg=dict(type='GloVe'))) ema = False ema_factor = 0.999 use_fp16 = False seed = 6666 deterministic = True log_level = 'INFO' log_interval = 50 save_interval = -1 resume_from = None load_from = 'work_dir/refcoco/PropVG-refcoco.pth' finetune_from = None evaluate_interval = 1 start_evaluate_epoch = 0 start_save_checkpoint = 20 max_token = 20 img_size = 384 patch_size = 16 model = dict( type='MIXRefUniModel_OMG', vis_enc=dict( type='BEIT3', img_size=384, patch_size=16, vit_type='base', drop_path_rate=0.1, vocab_size=64010, freeze_layer=-1, vision_embed_proj_interpolate=False, pretrain='pretrain_weights/beit3_base_patch16_224.zip'), lan_enc=None, fusion=None, head=dict( type='REFHead', input_channels=768, hidden_channels=256, num_queries=20, detr_loss=dict( criterion=dict(loss_class=1.0, loss_bbox=5.0, loss_giou=2.0), matcher=dict(cost_class=1.0, cost_bbox=5.0, cost_giou=2.0)), loss_weight=dict( mask=dict(dice=1.0, bce=1.0, nt=0.2, neg=0), bbox=0.1, allbbox=0.1, refer=1.0), MTD=dict(K=100)), post_params=dict( score_weighted=False, mask_threshold=0.5, score_threshold=0.7, with_nms=False, with_mask=True), process_visual=True, visualize_params=dict(row_columns=(4, 5)), visual_mode='test') grad_norm_clip = 0.15 lr = 0.0005 optimizer_config = dict( type='Adam', lr=0.0005, lr_vis_enc=5e-05, lr_lan_enc=0.0005, betas=(0.9, 0.98), eps=1e-09, weight_decay=0, amsgrad=True) scheduler_config = dict( type='MultiStepLRWarmUp', warmup_epochs=1, decay_steps=[21, 27], decay_ratio=0.1, max_epoch=30) launcher = 'pytorch' distributed = True rank = 0 world_size = 4 2025-07-07 11:04:11,542 - PropVG - INFO - RefCOCOUNC-val size: 10834 2025-07-07 11:04:17,084 - PropVG - INFO - RefCOCOUNC-testA size: 5657 2025-07-07 11:04:22,843 - PropVG - INFO - RefCOCOUNC-testB size: 5095 2025-07-07 11:04:28,381 - PropVG - INFO - loaded checkpoint from work_dir/refcoco/PropVG-refcoco.pth 2025-07-07 11:04:28,382 - PropVG - INFO - PropVG - evaluating set val 2025-07-07 11:06:19,535 - PropVG - INFO - ------------ validate ------------ time: 111.15, DetACC: 88.95, mIoU: 77.98, oIoU: 76.79, MaskACC@0.5-0.9: [89.14, 86.33, 81.66, 70.75, 36.42]DetACC@0.5-0.9: [88.95, 86.66, 82.62, 73.65, 47.82] 2025-07-07 11:06:21,284 - PropVG - INFO - PropVG - evaluating set testA 2025-07-07 11:07:29,418 - PropVG - INFO - ------------ validate ------------ time: 68.13, DetACC: 91.55, mIoU: 79.81, oIoU: 79.57, MaskACC@0.5-0.9: [91.66, 89.84, 85.42, 73.96, 36.22]DetACC@0.5-0.9: [91.55, 89.95, 85.94, 77.69, 51.57] 2025-07-07 11:07:30,844 - PropVG - INFO - PropVG - evaluating set testB 2025-07-07 11:08:36,434 - PropVG - INFO - ------------ validate ------------ time: 65.59, DetACC: 85.73, mIoU: 75.28, oIoU: 73.68, MaskACC@0.5-0.9: [84.95, 81.26, 76.06, 65.64, 38.81]DetACC@0.5-0.9: [85.73, 82.06, 76.20, 66.88, 41.97] 2025-07-07 11:08:37,918 - PropVG - INFO - sucessfully save the results to work_dir/refcoco/refer_output_thr0.7_no-nms_no-sw_0.5_100.xlsx !!!