How to get tflite models of a size similar to the project on Android (20 MB) ?

[carlosparra1998]

Hello everyone.

I need to get a tflite model from the weight obtained with the workouts (.weight), but this weight occupies about 200 MB, so when transforming this into tflite it occupies 200 MB, and then I need a much less capacity, like the original that occupies only 20 MB.

Thanks.

[mkj98]

the demo model used in the android app is YOLOv4-tiny model

1. check this guide to train your model link

2. after generating your whites file save the model with this command python save_model.py --weights ./data/yolov4-tiny.weights --output ./checkpoints/yolov4-tiny-416 --input_size 416 --model yolov4 --framework tflite --tiny

3. convert your model to tflite using this command: python convert_tflite.py --weights ./checkpoints/yolov4-tiny-416 --output ./checkpoints/yolov4-tiny-416.tflite

4. copy model to android assets folder.

5. change the model file name on MainActvity and DetectorActivity.

6. open YoloV4Classifier.java and change isTiny to true