tianzhongsong/tensorflow-quantization-test

Tensorflow quantization (float32-->int8) inference test

tensorflow-quantization-test's Issues

have you done pointnet?

I found Pointnet is in todo list.
Pointnet is 3d cnn, I am not sure it could be quantization very well.
Have you tried about this?

Could keras model use the convert weights(.h5)?

After using 'convert.py', the weights are int8, could i load it for keras model or just tensorflow model?

Is it faster than the not quantized version?

Hi, you show your result of accuracy, that's impressive! But I am wondering whether it is faster than the not quantized version. As you know, the conv operation is still conduct on float32 numbers. Thank you.