Part Number: TDA2EG-17
I'm following pytorch-jacinto-ai-devkit's instruction on installing the package, however, I encountered some problems:
(I'm building the kit in docker containers)
I started with a clean ubuntu:18.04 image, after installing conda (it installs python 3.7 automatically) and cloning the repo, then ran the ./setup.sh. for installation, I got an error by running the /scripts/train_classification_main.py:
raise AssertionError("Torch not compiled with CUDA enabled")
I also tried using pytorch's pytoch/pytorch:1.4-cuda10.1-cudnn7-devel image (official image by Pytorch), followed the same processes. I also got the error by running the /scripts/train_classification_main.py:
ImportError: cannot import name 'PILLOW_VERSION' from 'PIL'
Regardless of using the python script as suggested or not, error still occurred:
python ./scripts/train_classification_main.py --dataset_name cifar100_classification --model_name mobilenetv2_tv_x1 --data_path ./data/datasets/cifar100_classification --img_resize 32 --img_crop 32 --rand_scale 0.5 1.0
or simply:
./train_classification_main.py
Would you please check on the pytorch-jacinto-ai-devkil/setup.sh , making sure it's correct or I missed any installation steps which lead to the errors.
Also, ever consider release the Dockerfile and pre-build docker image which will significantly reduce this kinds of issues? Thank you.
