sovit-123 / fasterrcnn-pytorch-training-pipeline

PyTorch Faster R-CNN Object Detection on Custom Dataset
MIT License
223 stars 77 forks source link

can you add option to save the output to google drive directly #92

Closed VYRION-Ai closed 1 year ago

VYRION-Ai commented 1 year ago

thank you for your great work, when training on Collab( most of us do that ) the output saved in the Collab , please can you edit code to make the save to be the dir we write in the --name , so it take it as it , if i choose the google drive it go directly to drive , because training takes a lot of time , and many times session closed before saving to drive and i have to stope training to copy the output to drive

sovit-123 commented 1 year ago

@VYRION-Ai Thanks for the suggestion. However, there are a few issues with that. To save to Google Drive, you need to connect it first which asks for credentials. That cannot be done automatically. What I can do is add another argument that can take user path directly to save to a directory instead of outputs directory. But for that also, the user will have to connect to Google Drive first and then provide the custom directory path to the Drive.

I hope this makes sense. Let me know your thoughts.

VYRION-Ai commented 1 year ago

@sovit-123 yes connection is not done automatically. most of us load the dataset from google drive so we have to make connection , add new argument will be great , the important thing is to save automatically to drive . thank you @sovit-123 , this will be great achievement to us and will save us a lot

sovit-123 commented 1 year ago

Sure. I will try to push the update as soon as possible.

sovit-123 commented 1 year ago

@VYRION-Ai Now, you can give path to any custom directory including Google Drive using --project-dir command. Example: python train.py --data data_configs/custom_data.yaml --epochs 2 --model fasterrcnn_resnet50_fpn_v2 --project-dir "/content/drive/MyDrive/Colab Notebooks/fasterrcnn_repository/custom_training" --batch 4

Please pull the latest code.

VYRION-Ai commented 1 year ago

@sovit-123 thank you very much

sovit-123 commented 1 year ago

Welcome.