Sovit Ranjan Rath

@uysalfurkan Try with smaller batch size, in my opinion. Faster RCNN models are generally larger YOLOv5 models. From my experience, Faster RCNN models give good results with batch size 4...

Thank you for the feedback. @uysalfurkan I CSV file with mAP metric is already saved. Let me know what other things you want to track in the CSV file. I...

@uysalfurkan Hi, Some of the things like an opt.yaml file containing all the hyperparameters and model names are already saved to the resulting directory. I will try to add the...

@uysalfurkan I have also updated WandB to plot everything according to epoch instead of iterations which is a bit easier to interpret.

@uysalfurkan The CSV file update has been made. It's slightly different as of now compared to what you are asking. It has four losses, mAP @ 0.5 and mAP @...

@uysalfurkan I am not yet analyzing the validation values. It will take some time. It requires modification in the validation function inside the engine.py script. This is because Faster RCNN...

@uysalfurkan mAP is a validation metric already. In object detection, we calculate mAP on the validation dataset only which is the case with this code base as well. I hope...

@uysalfurkan In object detection, you can check that a model is overfitting if the mAP starts decreasing instead of increasing. mAP is always calculated on the validation dataset.

Hello @uysalfurkan That above warning says that are there 2 cores in the CPU but you are trying to use 4 cores (4 workers). I think this is a common...

@uysalfurkan Hello, that would require changing the Pycocotools code. But at the moment, I cannot say for sure at which place we need to change the code.