search

Z-Xiong / LightTrack-rknn

RKNN version demo of [CVPR21] LightTrack: Finding Lightweight Neural Network for Object Tracking via One-Shot Architecture Search
3 stars 0 forks source link

The rknn model is quantisized with int8 or just fp16? #2

Open SyGoing opened 1 year ago

SyGoing commented 1 year ago

Hello,Thank you for your wonderful work. The rknn model you converted with int8 or fp16? Since the models consist of backbone and neck_head, the neck_head's int8 convert is possible?

Z-Xiong commented 1 year ago

Thank you for your attention. I'm sorry I didn't see your issues before. I used defalut hybrid quantization(int8 and float32) on rv1109, because rv1109 don't support fp16. If convert neck_head to int8, I think precision will be lost.