2025-07-07 11:46:17,817 - PropVG - INFO - dataset = 'RRefCOCO'
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='LoadImageAnnotationsFromFileGRES_TO',
        max_token=50,
        with_mask=True,
        with_bbox=True,
        dataset='RRefCOCO',
        use_token_type='beit3',
        refer_file='./data/seqtr_type/annotations/rrefcoco/allobj.json',
        object_area_filter=100,
        object_area_rate_filter=[0.05, 0.8]),
    dict(type='Resize', img_scale=(320, 320), 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', 'gt_mask_parts_rle'
        ],
        meta_keys=[
            'filename', 'expression', 'ori_shape', 'img_shape', 'pad_shape',
            'scale_factor', 'gt_ori_mask', 'target', 'empty',
            'refer_target_index', 'tokenized_words'
        ])
]
val_pipeline = [
    dict(
        type='LoadImageAnnotationsFromFileGRES_TO',
        max_token=50,
        with_mask=True,
        with_bbox=True,
        dataset='RRefCOCO',
        use_token_type='beit3',
        refer_file='./data/seqtr_type/annotations/rrefcoco/allobj.json',
        object_area_filter=100,
        object_area_rate_filter=[0.05, 0.8]),
    dict(type='Resize', img_scale=(320, 320), 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', 'gt_mask_parts_rle'
        ],
        meta_keys=[
            'filename', 'expression', 'ori_shape', 'img_shape', 'pad_shape',
            'scale_factor', 'gt_ori_mask', 'target', 'empty',
            'refer_target_index', 'tokenized_words'
        ])
]
test_pipeline = [
    dict(
        type='LoadImageAnnotationsFromFile',
        max_token=20,
        with_bbox=True,
        dataset='RRefCOCO'),
    dict(type='Resize', img_scale=(512, 512), keep_ratio=False),
    dict(
        type='Normalize',
        mean=[123.675, 116.28, 103.53],
        std=[58.395, 57.12, 57.375]),
    dict(type='Pad', size_divisor=32),
    dict(type='DefaultFormatBundle'),
    dict(type='CollectData', keys=['img', 'ref_expr_inds', 'gt_bbox'])
]
word_emb_cfg = dict(type='GloVe')
data = dict(
    samples_per_gpu=16,
    workers_per_gpu=4,
    train=dict(
        type='RRefCOCO',
        which_set='train',
        img_source=['coco'],
        annsfile='./data/seqtr_type/annotations/rrefcoco/instance_withid.json',
        imgsfile='./data/seqtr_type/images/mscoco/train2014',
        pipeline=[
            dict(
                type='LoadImageAnnotationsFromFileGRES_TO',
                max_token=50,
                with_mask=True,
                with_bbox=True,
                dataset='RRefCOCO',
                use_token_type='beit3',
                refer_file='./data/seqtr_type/annotations/rrefcoco/allobj.json',
                object_area_filter=100,
                object_area_rate_filter=[0.05, 0.8]),
            dict(type='Resize', img_scale=(320, 320), 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', 'gt_mask_parts_rle'
                ],
                meta_keys=[
                    'filename', 'expression', 'ori_shape', 'img_shape',
                    'pad_shape', 'scale_factor', 'gt_ori_mask', 'target',
                    'empty', 'refer_target_index', 'tokenized_words'
                ])
        ],
        word_emb_cfg=dict(type='GloVe')),
    val_rrefcoco=dict(
        type='RRefCOCO',
        which_set='val_rrefcoco',
        img_source=['coco'],
        annsfile='./data/seqtr_type/annotations/rrefcoco/instance_withid.json',
        imgsfile='./data/seqtr_type/images/mscoco/train2014',
        pipeline=[
            dict(
                type='LoadImageAnnotationsFromFileGRES_TO',
                max_token=50,
                with_mask=True,
                with_bbox=True,
                dataset='RRefCOCO',
                use_token_type='beit3',
                refer_file='./data/seqtr_type/annotations/rrefcoco/allobj.json',
                object_area_filter=100,
                object_area_rate_filter=[0.05, 0.8]),
            dict(type='Resize', img_scale=(320, 320), 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', 'gt_mask_parts_rle'
                ],
                meta_keys=[
                    'filename', 'expression', 'ori_shape', 'img_shape',
                    'pad_shape', 'scale_factor', 'gt_ori_mask', 'target',
                    'empty', 'refer_target_index', 'tokenized_words'
                ])
        ],
        word_emb_cfg=dict(type='GloVe')),
    val_rrefcocoplus=dict(
        type='RRefCOCO',
        which_set='val_rrefcoco+',
        img_source=['coco'],
        annsfile='./data/seqtr_type/annotations/rrefcoco/instance_withid.json',
        imgsfile='./data/seqtr_type/images/mscoco/train2014',
        pipeline=[
            dict(
                type='LoadImageAnnotationsFromFileGRES_TO',
                max_token=50,
                with_mask=True,
                with_bbox=True,
                dataset='RRefCOCO',
                use_token_type='beit3',
                refer_file='./data/seqtr_type/annotations/rrefcoco/allobj.json',
                object_area_filter=100,
                object_area_rate_filter=[0.05, 0.8]),
            dict(type='Resize', img_scale=(320, 320), 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', 'gt_mask_parts_rle'
                ],
                meta_keys=[
                    'filename', 'expression', 'ori_shape', 'img_shape',
                    'pad_shape', 'scale_factor', 'gt_ori_mask', 'target',
                    'empty', 'refer_target_index', 'tokenized_words'
                ])
        ],
        word_emb_cfg=dict(type='GloVe')),
    val_rrefcocog=dict(
        type='RRefCOCO',
        which_set='val_rrefcocog',
        img_source=['coco'],
        annsfile='./data/seqtr_type/annotations/rrefcoco/instance_withid.json',
        imgsfile='./data/seqtr_type/images/mscoco/train2014',
        pipeline=[
            dict(
                type='LoadImageAnnotationsFromFileGRES_TO',
                max_token=50,
                with_mask=True,
                with_bbox=True,
                dataset='RRefCOCO',
                use_token_type='beit3',
                refer_file='./data/seqtr_type/annotations/rrefcoco/allobj.json',
                object_area_filter=100,
                object_area_rate_filter=[0.05, 0.8]),
            dict(type='Resize', img_scale=(320, 320), 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', 'gt_mask_parts_rle'
                ],
                meta_keys=[
                    'filename', 'expression', 'ori_shape', 'img_shape',
                    'pad_shape', 'scale_factor', 'gt_ori_mask', 'target',
                    'empty', 'refer_target_index', 'tokenized_words'
                ])
        ],
        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/rrefcoco/PropVG-rrefcoco.pth'
finetune_from = None
evaluate_interval = 1
start_evaluate_epoch = 0
start_save_checkpoint = 9
max_token = 50
img_size = 320
patch_size = 16
num_queries = 20
model = dict(
    type='MIXRefUniModel_OMG',
    vis_enc=dict(
        type='BEIT3',
        img_size=320,
        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='GTMHead',
        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=250)),
    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=[7, 11],
    decay_ratio=0.1,
    max_epoch=12)
launcher = 'none'
distributed = False
rank = 0
world_size = 1

2025-07-07 11:46:34,374 - PropVG - INFO - RRefCOCO-val_rrefcoco size: 52229
2025-07-07 11:46:53,442 - PropVG - INFO - RRefCOCO-val_rrefcoco+ size: 49620
2025-07-07 11:47:11,525 - PropVG - INFO - RRefCOCO-val_rrefcocog size: 33960
2025-07-07 11:47:16,069 - PropVG - INFO - loaded checkpoint from work_dir/rrefcoco/PropVG-rrefcoco.pth

2025-07-07 11:47:16,070 - PropVG - INFO - PropVG - evaluating set val_rrefcoco
2025-07-07 11:58:15,741 - PropVG - INFO - ------------ validate ------------time: 659.65, mIoU: 75.86, oIoU: 76.87, mRR: 93.03, rIoU: 62.91
2025-07-07 11:58:18,322 - PropVG - INFO - PropVG - evaluating set val_rrefcoco+
2025-07-07 12:07:56,811 - PropVG - INFO - ------------ validate ------------time: 578.47, mIoU: 69.39, oIoU: 69.17, mRR: 94.96, rIoU: 59.44
2025-07-07 12:07:58,975 - PropVG - INFO - PropVG - evaluating set val_rrefcocog
2025-07-07 12:14:35,849 - PropVG - INFO - ------------ validate ------------time: 396.86, mIoU: 69.20, oIoU: 70.13, mRR: 93.85, rIoU: 56.17
2025-07-07 12:14:37,866 - PropVG - INFO - sucessfully save the results to work_dir/rrefcoco/refer_output_thr0.7_no-nms_no-sw_0.5_250.xlsx !!!