Inference error

[davidvct]

1. I used this command for inference but encountered issue. Anyone knows how to fix this?

• command: python launch.py --n_GPUs 1 main.py --batch_size 8 --precision single
• error :
  [W socket.cpp:401] [c10d] The server socket has failed to bind to [::]:8023 (errno: 98 - Address already in use). [W socket.cpp:401] [c10d] The server socket has failed to bind to workstation2:8023 (errno: 98 - Address already in use). [E socket.cpp:435] [c10d] The server socket has failed to listen on any local network address.

2. Another question is, how do I specify which model to use for inference?

[SeungjunNah]

1. Are you launching many jobs from a single machine? Use different master ports per job.
   https://github.com/SeungjunNah/DeepDeblur-PyTorch/blob/master/src/option.py#L33

2. Please provide more details.

[davidvct]

1. Changed the port and it works. Thanks!

2. I was asking how to perform inference on test datasets with specific saved model. Now I managed to get the prediction running, using the below command:
   python launch.py --n_GPUs 1 main.py --save_dir 2024-04-01_14-13-08 --do_train False --do_validate False --start_epoch 270 --load_epoch 270

   But there are two issues I encountered:

   • the inference only perform images on some images. I have 60 test images in a test folder but only 6 were predicted.

   • after finished prediction with model-270.pt, the script will proceed predict with model-280.pt. It is not a big issue, but something to consider for future improvement.

[SeungjunNah]

2. Please refer to Usage examples - Example commands

# save all of the evaluation results
python main.py --n_GPUs 1 --batch_size 8 --dataset GOPRO_Large --save_results all

Please refer to args.end_epoch and see how it is used in main.py.