Sterling Nesbitt

Haven't tried on Android. But I did transfer learning (started with pretrained weights but had custom dataset), converted to tflite (both regular and int8 (int8 required some hacks)), and ran...

Try: python convert_tflite.py --weights ./checkpoints/yolov4 And change the load function to: model.load_weights(FLAGS.weights) https://github.com/hunglc007/tensorflow-yolov4-tflite/issues/68

tflite doesn't support load_weights. You have to use tf.lite.Interpreter

Put the path of your model here. I used h5 format. It should work with file format too https://github.com/sterlingrpi/ssd_tflite_tpu/blob/c693f0b506f18091787bea8b1c33a94936b00bb5/tflite_convert.py#L10

What class has you doing a custom trained SSD on a TPU?

NM. I thought you were asking about a different repo

@zorrobyte same issue here. Changing /sys/class/thermal/thermal_zone0/trip_point_4_temp doesn't last after a reboot. Any solution? edit, wanted to tag this related thread and @lleonid since he seems to know a lot https://github.com/f0cal/google-coral/issues/50

Have you converted the model to int8 tflite and used the tpu compiler? This will reduce the size. I am running on TPU and raspberry pi. So I will be...

@javimg140 the trouble I am seeing with converting this to TPU is the model is defined as a subclass, which is not supported by tflite. I am working on defining...

@javimg140 I got the model defined using functional API and it converts to tflite int8 and TPU. Now my problem is training on my own data doesn't yield good results....