Sovit Ranjan Rath
Sovit Ranjan Rath
From a first look, it seemed like an internal transform issue of the model. However, the transforms are the same as the ResNet backbone ones. So, I still need to...
For tiny objects, I recommend Faster RCNN ResNet50 FPN V2.
@samahwaleed ResNet101 model is not yet added. Will add that for sure if that is a requirement. Regarding precision and recall. You can check precision and recall, you can visualize...
Sure, I would add the model by tomorrow. After training the model, use the `eval.py` script to check the metrics. Example command: `python eval.py --weights outputs/training/custom_training/best_model.pth --config data_configs/custom_data.yaml --model fasterrcnn_resnet50_fpn_v2...
Right now, the library uses Torchmetrics to show class-wise mAP. I will try to add classwise recall and precision. But it may take some time.
I can add recall per class but not sure about the F1 score. Before that, I will add the ResNet101 model.
@samahwaleed I have added the ResNet101 model.
@GeorgePearse Thanks for the support. And you are right in a big way. The fasterrcnn_resnet50_fpn_v2 is one of the safest model choices. It really does very well on a variety...
@EmmaVanPuyenbr Hi. Does it complete entirely one epoch (training + validation)? Or the error happens on the first epoch only after the training loop completes?
Can you check your GPU memory usage once while training? After the training loop, there will be a slight surge in the GPU memory usage when the validation loop starts....