zjykzj / crnn-ctc Goto Github PK

pip install -r requirements.txt

$ python -m torch.distributed.run --nproc_per_node 4 --master_port 32512 train_emist.py --device 0,1,2,3 ../datasets/EMNIST/ runs/emnist_ddp/

$ python -m torch.distributed.run --nproc_per_node 4 --master_port 32512 train_plate.py --device 0,1,2,3 ../datasets/git_plate/CCPD_CRPD_OTHER_ALL/ ../datasets/git_plate/val_verify/ runs/plate_ddp/

$ python eval_emnist.py runs/emnist_ddp/crnn-emnist-e100.pth ../datasets/EMNIST/
args: Namespace(pretrained='runs/emnist_ddp/crnn-emnist-e100.pth', val_root='../datasets/EMNIST/')
Loading CRNN pretrained: runs/emnist_ddp/crnn-emnist-e100.pth
Batch:62 ACC:100.000: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 63/63 [00:02<00:00, 30.82it/s]
ACC:95.100

$ python eval_plate.py runs/plate_ddp/crnn-plate-e100.pth ../datasets/git_plate/val_verify/
args: Namespace(pretrained='runs/plate_ddp/crnn-plate-e100.pth', val_root='../datasets/git_plate/val_verify/')
Loading CRNN pretrained: runs/plate_ddp/crnn-plate-e100.pth
Load test data: 2014
Batch:62 ACC:96.667: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 63/63 [00:02<00:00, 22.51it/s]
ACC:97.319

$ python predict_emnist.py runs/emnist_ddp/crnn-emnist-e100.pth ../datasets/EMNIST/ runs/
args: Namespace(pretrained='runs/emnist_ddp/crnn-emnist-e100.pth', save_dir='runs/', val_root='../datasets/EMNIST/')
Loading CRNN pretrained: runs/emnist_ddp/crnn-emnist-e100.pth
Label: [6 5 3 8 6] Pred: [6 5 3 8 6]
Label: [2 1 4 3 0] Pred: [2 1 4 3 0]
Label: [8 5 3 3 5] Pred: [8 5 3 3 5]
Label: [0 7 9 4 9] Pred: [0 7 9 4 9]
Label: [6 1 7 2 0] Pred: [6 1 7 2 0]
Label: [8 8 9 9 5] Pred: [8 8 9 9 5]

$ python predict_plate.py runs/plate_ddp/crnn-plate-e100.pth ./assets/plate/宁A87J92_0.jpg runs/
args: Namespace(image_path='./assets/plate/宁A87J92_0.jpg', pretrained='runs/plate_ddp/crnn-plate-e100.pth', save_dir='runs/')
Loading CRNN pretrained: runs/plate_ddp/crnn-plate-e100.pth
Pred: 宁A87J92
$ python predict_plate.py runs/plate_ddp/crnn-plate-e100.pth ./assets/plate/川A3X7J1_0.jpg runs/
args: Namespace(image_path='./assets/plate/川A3X7J1_0.jpg', pretrained='runs/plate_ddp/crnn-plate-e100.pth', save_dir='runs/')
Loading CRNN pretrained: runs/plate_ddp/crnn-plate-e100.pth
Pred: 川A3X7J1