cnn

Results 7 issues of cnn

1.加载预训练模型时,如果自己的数据集分类数不是1000类,需要过滤不需要的layer 2.加载数据时默认减去mean值 - 如果加载预训练模型,需要将mean值设置跟预训练模型一致 - 如果不加载预训练模型,加载数据也要考虑归一化的问题

**Describe the bug** RT, what is the difference between mmdetection-to-tensorrt and torch2trt (https://github.com/NVIDIA-AI-IOT/torch2trt)?

pls modify the Dockerfile FROM tf-model-server as XXX/tf-model-server

fix code indent

RT, is it available that load triton OPs on ONNX model?

The jina version is jina>=3.12.0 on setup.py. But the tenosrrt.dockerfile jina version is "jina[standard]==${JINA_VERSION}". Is that right?

add actionCLIP