1. Download pyruntime.tgz from NAS server("ai/prebuilt/pyruntime.tgz")
This package provides a complete environment of python3.6, and also contains
modules of torch, torchvision, numpy and cvitek's runtime python wrapper.
2. Decompress pacakge to soc platform:
$ cd /mnt/data
$ tar -zxvf pyruntime.tgz ./
3. Setup environment variables.
$ export PATH=/mnt/data/python/bin/:/mnt/data/cvitek_tpu_sdk/bin:${PATH}
$ export LD_LIBRARY_PATH=/mnt/data/python/lib:/mnt/data/cvitek_tpu_sdk/lib:${LD_LIBRARY_PATH}
$ export PYTHON_EGG_CACHE=/mnt/data/python/.cache/
4. Run accuracy evaluation script:
$ ./eval_imagenet.py \
--dataset imagenet/img_val_extracted/ \
--cvimodel resnet50_bs4.cvimodel \
--count 1000 --batch_size 4 \
--image_resize_dims 256,256 \
--net_input_dims 224,224 \
--mean 104.01,116.67,122.68