3. t2t_vit inference

前言

对vit 进行fp16推理

参考链接：

https://github.com/open-mmlab/mmpretrain/tree/master/configs/t2t_vit

run code ：

https://mmclassification.readthedocs.io/en/latest/getting_started.html#inference-and-test-a-dataset

https://mmclassification.readthedocs.io/en/latest/tools/pytorch2onnx.html?highlight=onnx

run code and convert onnx

# run acc 
python3 tools/test.py configs/t2t_vit/t2t-vit-t-14_8xb64_in1k.py  ./t2t-vit-t-14_8xb64_in1k_20211220-f7378dd5.pth --metrics accuracy  --out result.pkl 2>&1 | tee test_run.log

# convert onnx , bs=1
pip install onnx  onnxsim
python3 tools/deployment/pytorch2onnx.py \
    configs/t2t_vit/t2t-vit-t-14_8xb64_in1k.py \
    --checkpoint ./t2t-vit-t-14_8xb64_in1k_20211220-f7378dd5.pth  \
    --output-file ./t2t-vit-t-14.onnx  2>&1 | tee test_onnx.log

# simplify
onnxsim t2t-vit-t-14.onnx  t2t-vit-t-14-