fix khwc usage (#352)

optimize tensorrt, better argmax impl

refine the logger setup

use torch2.10 to export onnx, and a bit refine of trt10 code (#349)

Update README.md

refine tensorrt, and use plugin (#340)

onnx (#336)

update to tensorrt 10.3 (#333)

ncnn (#288)

a bit fix of readme and small refinement (#282)

add int8 inference for trt-cpp (#272)

fix tensorrt cmakelist.txt (#264)

move preprocessing to server side as a backend, and add http client (#261)

write argmax (#254)

update training command (#253)