search

Zhen-Dong / HAWQ

Quantization library for PyTorch. Support low-precision and mixed-precision quantization, with hardware implementation through TVM.
MIT License
406 stars 83 forks source link

[Feature request] Support for 3d models #9

Open dmenig opened 3 years ago

dmenig commented 3 years ago

It'd be very interesting, especially since 3d models are part of torchvision model zoo.

Zhen-Dong commented 3 years ago

Hi, thanks for your interest in the repo. Currently, my roadmap is to first add more network architectures (sadly they are 2D models), and then train/quantize more checkpoints to strengthen the model zoo. As a result, I may get no extra time in the near future. However, the quantization tools are general, so to support 3D models the main point is to set up the normal training pipeline for 3D models and tasks. I believe it should be a feasible thing to do.