2025-07-07 11:14:04,236 - PropVG - INFO - dataset = 'RefCOCOgUMD'
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='RefCOCOgUMD',
        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='RefCOCOgUMD',
        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='RefCOCOgUMD',
        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='RefCOCOgUMD',
        which_set='train',
        img_source=['coco'],
        annsfile=
        './data/seqtr_type/annotations/refcocog-umd/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='RefCOCOgUMD',
                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='RefCOCOgUMD',
        which_set='val',
        img_source=['coco'],
        annsfile=
        './data/seqtr_type/annotations/refcocog-umd/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='RefCOCOgUMD',
                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')),
    test=dict(
        type='RefCOCOgUMD',
        which_set='test',
        img_source=['coco'],
        annsfile=
        './data/seqtr_type/annotations/refcocog-umd/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='RefCOCOgUMD',
                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/refcocog/PropVG-refcocog.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 = 1

2025-07-07 11:14:09,303 - PropVG - INFO - RefCOCOg-val size: 4896
2025-07-07 11:14:14,811 - PropVG - INFO - RefCOCOg-test size: 9602
2025-07-07 11:14:19,468 - PropVG - INFO - loaded checkpoint from work_dir/refcocog/PropVG-refcocog.pth

2025-07-07 11:14:19,479 - PropVG - INFO - PropVG - evaluating set val
2025-07-07 11:16:13,025 - PropVG - INFO - ------------ validate ------------  time: 113.54, DetACC: 83.50, mIoU: 71.34, oIoU: 69.30, MaskACC@0.5-0.9: [81.19, 77.33, 71.51,  60.15,  30.78]DetACC@0.5-0.9: [83.50, 80.09, 75.41,  66.07,  40.54]
2025-07-07 11:16:15,090 - PropVG - INFO - PropVG - evaluating set test
2025-07-07 11:19:29,251 - PropVG - INFO - ------------ validate ------------  time: 194.16, DetACC: 84.44, mIoU: 72.10, oIoU: 70.53, MaskACC@0.5-0.9: [82.53, 78.47, 72.66,  61.23,  30.31]DetACC@0.5-0.9: [84.44, 81.32, 76.33,  67.14,  42.69]
2025-07-07 11:19:31,176 - PropVG - INFO - sucessfully save the results to work_dir/refcocog/refer_output_thr0.7_no-nms_no-sw_0.5_100.xlsx !!!