wangermeng2021

Sorry for late replay, It's seem that batch_boxes[batch_index][box_index][4].astype(np.int32) exceed max limit 5+32(num-classes), because your num-classes is 32, but one of your classes included in your coco dataset is 39, so...

Could you share your command line? what's your "model-type"? p5? what's your "--multi-scale"? It's seem that image size is not divided by 32(p5) / 64(p6) / 128(p7) .

Thanks for the feedback, My test environment is Tensorflow2.3. I will check it. It should generate new different tensor name in every loop

> I managed to fix the problem! > > The issue lies in the way the model is exported. More precisely, [line 294](https://github.com/wangermeng2021/Scaled-YOLOv4-tensorflow2/blob/8eab3fb40d46e1aeb030a326fdcf889fae228217/train.py#L294) in `train.py` calls the Keras API method...

tensorflow 2.3, numpy1.19 python3.6, pycocotools2.0