Sovit Ranjan Rath

Did you try training again. That mAP is the validation mAP at 67th epoch. And that is the final best weight. No other models were trained after that. For the...

So, after 67 epochs, the mode is starting to overfit. That's why I chose that model. And I will correct the traffic signs to traffic lights. Thanks for pointing that...

Yes, on the validation dataset.

Hello. Please note that the dataset is one folder back in the `input` directory. Therefore, it is `../input/rest_of_the_path`. Can you please make sure it is the same case in your...

Can you try giving relative path like the code expects? The code in this repo does not expect Kaggle environment but local system. So, having the data one folder back...

@MheadHero There are a few ways. But there are other important things that you may want to consider before moving into the validation loss path. One thing is that in...

Hi, I think that your classes names or the number of classes (NC) in the data YAML file are wrong. Can you recheck it or paste your dataset YAML file...

@ilpapds As I remember it it was to work around some GPU memory issue. If you can train with higher batch size, surely go for it.

@ilpapds If you asking about the Darknet CFG models, yes, they are the implementation of the original models. In fact, the YOLOv4 Darknet repository and the paper were developed by...

@TerrafYassin You can take a look at this model => https://github.com/sovit-123/fastercnn-pytorch-training-pipeline/blob/main/models/fasterrcnn_mini_darknet_mini_head.py It contains custom head and Predictor. You will get an idea of how to accomplish what you want.