search

hunglc007 / tensorflow-yolov4-tflite

YOLOv4, YOLOv4-tiny, YOLOv3, YOLOv3-tiny Implemented in Tensorflow 2.0, Android. Convert YOLO v4 .weights tensorflow, tensorrt and tflite
https://github.com/hunglc007/tensorflow-yolov4-tflite
MIT License
2.23k stars 1.24k forks source link

Model weights wont converge #379

Open ghanbarzadeh opened 3 years ago

ghanbarzadeh commented 3 years ago

I'm trying to train the model on PASCAL 2012, however, the model doesn't seem to converge with my settings (Batch size: 2, Rezolution: 416). This is the highest batch size I could go without my GPU memory running out. I was wondering if anybody has had success training with these parameters?

ahmethamzakaya commented 3 years ago

Yes i also used small batch on my custom data since my gpu has 4gb vram, but after around 7000 iterations model converged to get above 85% maP for each class . I recommend to train your model in darknet using " ./darknet detector train command", then convert weights to tensorflow using this repo.

Also you can refer here for some tricks: https://github.com/AlexeyAB/darknet#how-to-improve-object-detection