The RegionYolo layer is not yet implemented.

[pinto0309]

The RegionYolo layer is not yet implemented.

[PINTO0309]

Uh... For the smaller model, I use the ultra-light version called YoloV4-tiny. The 250MB size YoloV4 is the full size YoloV4. As expected, it is impossible to compress a 250MB model to 2-3MB.
https://github.com/PINTO0309/PINTO_model_zoo/tree/master/046_yolov4-tiny

I used the following repository.
https://github.com/bubbliiiing/yolov4-tiny-keras.git

[semihstp]

Okey. I will try to do training on yolov4-tiny model and then i will convert .tflite. Thanks for your advice :)

[PINTO0309]

It is not practical to reimplement RegionYolo in TensorFlow. So let me ask you a few questions in order to find a solution.

1. Is it a requirement to train with Darknet? Many YoloV4 implementations of frameworks other than Darknet, such as PyTorch and TensorFlow, have been released.
2. Can you try DW2TF? https://github.com/jinyu121/DW2TF
3. Can you try tensorflow-yolov4-tflite? https://github.com/hunglc007/tensorflow-yolov4-tflite
4. Does your YoloV4 model have any special customization against the official model?

[semihstp]

Yes i have to train with Yolov4. The project about an android detection app based on yolov4 and Tensorflow Lite.
I have tried the 2 and 3 repositories but converted tflite files has almost 250MB. So i cant use them on the app. When i saw your repo and gifs, i saw that your tflite size 3-5MB like i want to convert.

I didint any change yolov4 model. Just train my own dataset and classes. I have 32 classes in this case.