models/config.py

work_dir = 'records/guoshoucai_auto_gen_ps_with_tianchi_psccnet_baseline_dct_balance_scale_0_05_1_0_15_epochs_cls_weight_1_5_more_negs_seed_4567'
dataset_type = 'MaskSegDatasetv2'
img_norm_cfg = dict(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
input_size = (512, 512)
train_pre_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(
        type='LoadAnnotations', binary=True, train=True, img_label_binary=True)
]
train_post_pipeline = [
    dict(type='SimpleResize', size=(512, 512)),
    dict(type='RandomFlip', prob=0.5),
    dict(
        type='Normalizev2',
        mean=[0.485, 0.456, 0.406],
        std=[0.229, 0.224, 0.225]),
    dict(type='DefaultFormatBundle'),
    dict(type='Collect', keys=['img', 'gt_semantic_seg', 'img_label'])
]
test_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(type='SimpleResize', size=(512, 512)),
    dict(
        type='Normalizev2',
        mean=[0.485, 0.456, 0.406],
        std=[0.229, 0.224, 0.225]),
    dict(type='ImageToTensor', keys=['img']),
    dict(type='Collect', keys=['img'])
]
data = dict(
    samples_per_gpu=1,
    workers_per_gpu=4,
    train=dict(
        type='MaskSegDatasetv2',
        data_root='/mnt/disk1/data/image_forgery/text_forgery',
        ann_path='guoshoucai_auto_gen_ps_with_tianchi_1.txt',
        pipeline=[[{
            'type': 'LoadImageFromFile'
        }, {
            'type': 'LoadAnnotations',
            'binary': True,
            'train': True,
            'img_label_binary': True
        }],
                  [{
                      'type': 'SimpleResize',
                      'size': (512, 512)
                  }, {
                      'type': 'RandomFlip',
                      'prob': 0.5
                  }, {
                      'type': 'Normalizev2',
                      'mean': [0.485, 0.456, 0.406],
                      'std': [0.229, 0.224, 0.225]
                  }, {
                      'type': 'DefaultFormatBundle'
                  }, {
                      'type': 'Collect',
                      'keys': ['img', 'gt_semantic_seg', 'img_label']
                  }]]),
    val=[
        dict(
            type='MaskSegDatasetv2',
            data_root=
            '/mnt/disk1/data/image_forgery/text_forgery/guoshoucai_auto_gen/test_forged_with_ps',
            ann_path='test_1.txt',
            test_mode=True,
            pipeline=[
                dict(type='LoadImageFromFile'),
                dict(type='SimpleResize', size=(512, 512)),
                dict(
                    type='Normalizev2',
                    mean=[0.485, 0.456, 0.406],
                    std=[0.229, 0.224, 0.225]),
                dict(type='ImageToTensor', keys=['img']),
                dict(type='Collect', keys=['img'])
            ],
            dataset_name='guoshoucai_text',
            gt_seg_map_loader_cfg=dict(binary=True, img_label_binary=True)),
        dict(
            type='MaskSegDatasetv2',
            data_root=
            '/mnt/disk1/data/image_forgery/text_forgery/tianchi_text_forgory',
            ann_path='val.txt',
            test_mode=True,
            pipeline=[
                dict(type='LoadImageFromFile'),
                dict(type='SimpleResize', size=(512, 512)),
                dict(
                    type='Normalizev2',
                    mean=[0.485, 0.456, 0.406],
                    std=[0.229, 0.224, 0.225]),
                dict(type='ImageToTensor', keys=['img']),
                dict(type='Collect', keys=['img'])
            ],
            dataset_name='tianchi',
            gt_seg_map_loader_cfg=dict(binary=True, img_label_binary=True))
    ])
norm_cfg = dict(type='SyncBN', requires_grad=True)
model = dict(
    type='PSCCDetector',
    base_model=dict(
        type='PSCCNet',
        crop_size=(512, 512),
        pretrained=
        '/home/yangwu/.cache/torch/checkpoints/hrnet_w18_small_v2.pth'),
    train_cfg=dict(
        seg_loss=dict(type='BCELoss', reduction='none'),
        seg_loss_weights=(1.0, 1.0),
        mask_loss_weights=(1.0, 1.0, 1.0, 1.0),
        cls_loss=dict(
            type='CrossEntropyLoss',
            use_sigmoid=False,
            class_weight=(1.0, 1.0)),
        p_balance_scale=0.05,
        n_balance_scale=1.0),
    test_cfg=dict())
optimizer = dict(type='Adam', lr=0.0001, weight_decay=1e-05)
optimizer_config = dict()
lr_config = dict(policy='CosineAnnealing', min_lr=1e-07, by_epoch=False)
runner = dict(type='IterBasedRunner', max_iters=121960)
checkpoint_config = dict(by_epoch=False, interval=4065, max_keep_ckpts=1)
evaluation = dict(
    interval=4065,
    metric='mFscore',
    pre_eval=True,
    mean=False,
    thresh=0.5,
    img_thresh=0.5)
log_config = dict(
    interval=50,
    hooks=[
        dict(type='TextLoggerHook', by_epoch=False),
        dict(type='TensorboardLoggerHook')
    ])
ext_test_dataset = ['CASIA1']
dist_params = dict(backend='nccl')
log_level = 'INFO'
load_from = None
resume_from = None
workflow = [('train', 1)]
cudnn_benchmark = True
find_unused_parameters = False
auto_resume = False
gpu_ids = range(0, 4)