PSNR very low, testing with EDVR pretrained model on single gpu

[mach-ms]

My env :Windows 11, single RTX3090 GPU. mmediting version (mmedit.version): '0.15.1' I use the script tools/test.py to test the EDVR Large model download from: https://download.openmmlab.com/mmediting/restorers/edvr/edvrl_c128b40_8x8_lr2e-4_600k_reds4_20220104-4509865f.pth Dataset is REDS. Testing result on cmd is:

load checkpoint from local path: C:\...\edvrl_c128b40_8x8_lr2e-4_600k_reds4_20220104-4509865f.pth [>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 400/400, 1.6 task/s, elapsed: 257s, ETA: 0s Eval-PSNR: 10.717399841857048 Eval-SSIM: 0.30903949172869444

PSNR and SSIM are very low.

Attaching my config script edvrl_c128b40_8x8_lr2e-4_600k_reds4.py below:

edvrl_c128b40_8x8_lr2e-4_600k_reds4.py

exp_name = 'edvrl_c128b40_8x8_lr2e-4_600k_reds4'

model = dict( type='EDVR', generator=dict( type='EDVRNet', in_channels=3, out_channels=3, mid_channels=128, num_frames=5, deform_groups=8, num_blocks_extraction=5, num_blocks_reconstruction=40, center_frame_idx=2, with_tsa=True), pixel_loss=dict(type='CharbonnierLoss', loss_weight=1.0, reduction='sum'))

train_cfg = dict(tsa_iter=5000) test_cfg = dict(metrics=['PSNR', 'SSIM'], crop_border=0)

train_dataset_type = 'SRREDSDataset' val_dataset_type = 'SRREDSDataset' train_pipeline = [ dict(type='GenerateFrameIndices', interval_list=[1], frames_per_clip=99), dict(type='TemporalReverse', keys='lq_path', reverse_ratio=0), dict( type='LoadImageFromFileList', io_backend='disk', key='lq', flag='unchanged'), dict( type='LoadImageFromFileList', io_backend='disk', key='gt', flag='unchanged'), dict(type='RescaleToZeroOne', keys=['lq', 'gt']), dict( type='Normalize', keys=['lq', 'gt'], mean=[0, 0, 0], std=[1, 1, 1], to_rgb=True), dict(type='PairedRandomCrop', gt_patch_size=256), dict( type='Flip', keys=['lq', 'gt'], flip_ratio=0.5, direction='horizontal'), dict(type='Flip', keys=['lq', 'gt'], flip_ratio=0.5, direction='vertical'), dict(type='RandomTransposeHW', keys=['lq', 'gt'], transpose_ratio=0.5), dict(type='Collect', keys=['lq', 'gt'], meta_keys=['lq_path', 'gt_path']), dict(type='FramesToTensor', keys=['lq', 'gt']) ]

test_pipeline = [ dict(type='GenerateFrameIndiceswithPadding', padding='reflection_circle'), dict( type='LoadImageFromFileList', io_backend='disk', key='lq', flag='unchanged'), dict( type='LoadImageFromFileList', io_backend='disk', key='gt', flag='unchanged'), dict(type='RescaleToZeroOne', keys=['lq', 'gt']), dict( type='Normalize', keys=['lq', 'gt'], mean=[0, 0, 0], std=[1, 1, 1], to_rgb=True), dict( type='Collect', keys=['lq', 'gt'], meta_keys=['lq_path', 'gt_path', 'key']), dict(type='FramesToTensor', keys=['lq', 'gt']) ]

demo_pipeline = [ dict(type='GenerateSegmentIndices', interval_list=[1]), dict( type='LoadImageFromFileList', io_backend='disk', key='lq', channel_order='rgb'), dict(type='RescaleToZeroOne', keys=['lq']), dict(type='FramesToTensor', keys=['lq']), dict(type='Collect', keys=['lq'], meta_keys=['lq_path', 'key']) ]

data = dict( workers_per_gpu=8, train_dataloader=dict(samples_per_gpu=4, drop_last=True), val_dataloader=dict(samples_per_gpu=1), test_dataloader=dict(samples_per_gpu=1), train=dict( type='RepeatDataset', times=1000, dataset=dict( type=train_dataset_type, lq_folder='D:\DATASETS\REDS\train_sharp_bicubic\X4', gt_folder='D:\DATASETS\REDS\train_orig', ann_file='D:\DATASETS\REDS\meta_info_REDS_GT.txt', num_input_frames=5, pipeline=train_pipeline, scale=4, val_partition='REDS4', test_mode=False)), val=dict( type=val_dataset_type, lq_folder='D:\DATASETS\REDS\train_sharp_bicubic\X4', gt_folder='D:\DATASETS\REDS\train_orig', ann_file='D:\DATASETS\REDS\meta_info_REDS_GT.txt', num_input_frames=5, pipeline=test_pipeline, scale=4, val_partition='REDS4', test_mode=True), test=dict( type=val_dataset_type, lq_folder='D:\DATASETS\REDS\train_sharp_bicubic\X4', gt_folder='D:\DATASETS\REDS\train_orig', ann_file='D:\DATASETS\REDS\meta_info_REDS_GT.txt', num_input_frames=5, pipeline=test_pipeline, scale=4, val_partition='REDS4', test_mode=True), )

optimizers = dict(generator=dict(type='Adam', lr=2e-4, betas=(0.9, 0.999)))

total_iters = 600000 lr_config = dict( policy='CosineRestart', by_epoch=False, periods=[150000, 150000, 150000, 150000], restart_weights=[1, 0.5, 0.5, 0.5], min_lr=1e-7)

checkpoint_config = dict(interval=5000, save_optimizer=True, by_epoch=False)

evaluation = dict(interval=1000, save_image=True) log_config = dict( interval=100, hooks=[ dict(type='TextLoggerHook', by_epoch=False), # dict(type='TensorboardLoggerHook'), ]) visual_config = None

dist_params = dict(backend='nccl') log_level = 'INFO' work_dir = f'./work_dirs/{exp_name}' load_from = None # noqa: E501 resume_from = None workflow = [('train', 1)]