Training yolo on a custom dataset.

[makramhamad]

I am trying to overfit yolov4 model on a small dataset. I have followed instructions from the repo and changed annotations format as it is suggested by the repo. Dataset has only two classes it is ball and person. I make changes in cfg.yolov4 accordingly but when I run the training code, it does not train at all and shows following output after evaluation on validation dataset. please anyone guide me what am I doing wrong. P.S I have trained yolov5 on same dataset that goes great.

[ElHouas]

Hey @makramhamad,

I have the same issue when training for a custom dataset with 8 classes. Have you solved it?

Thanks

[makramhamad]

@ElHouas Hi, I solved it after increasing the dataset size. I was overfitting it on some 10-20 images but when i increased number of images results started improving with each epoch.

[ElHouas]

@makramhamad Thanks for your quick response. I would like to know if you made any additional changes to make it work using a custom dataset. I am still not availble to get better numbers even using a large dataset.

Thanks

[ingbeeedd]

This is the result of running with the original project when training with coco dataset.

[Wang-Zhenxing]

This is the result of running with the original project when training with coco dataset.

I have the same issue when training for a custom dataset. Have you solved it?