AI research relies much on learning codes written by other experienced researchers. However, many of open-source codes requires divergent environments. The environment configuration files inside are usually too strict to be compatible with each other. In this repo, we build a general conda environment that can cover as much repositories as possible.
| Perception | Generation | Vision&LLM | Others |
| Detectron2 Based Others | Diffusion Based GAN Based | ChatGPT Based |
Feel free to experiment with other repos, and pull requests.
- Set up a conda environment the same as us.
- Check if other repos can run directly. If not, try make minimal changes on existing packages.
- Provide short guidance. This is especially important if the pipeline under our env is different from the official tutorial.
git clone https://github.com/volgachen/Awesome-AI-Environment.git
cd Awesome-AI-Environment
conda env create -f conda_env.yaml
conda activate awesome_env
pip install -r requirements.txt
pip install -r req_no_deps.txt --no-deps
# install the latest version of detectron2
git clone https://github.com/facebookresearch/detectron2.git
python -m pip install -e detectron2
# install maskformer
git clone https://github.com/NVlabs/ODISE.git
pip install panopticapi@https://github.com/cocodataset/panopticapi/archive/master.zip
pip install ODISE/third_party/mask2former --no-deps
# we take the env above, the official one from https://github.com/facebookresearch/Mask2Former.git has not been validated
- Chinese users may refer here for faster installations with mirror source.
Detectron has been installed in an editable way. You may edit and import it following official instructions.
We consider Detrex as a code project, and would not link it to site-packages. Therefore, the following example codes only ensures that you can run detrex inside the folder.
git clone [email protected]:IDEA-Research/detrex.git
cd detrex
python setup.py build_ext --inplace
PYTHONPATH=/path/to/repo:$PYTHONPATH python tools/train_net.py --config-file ${CONFIG_FILE} --num-gpus ${NUM_GPUS}
We consider ODISE as a code project, and would not compile and link it to site-packages. Therefore, the following example codes only ensures that you can run ODISE inside the folder.
git clone [email protected]:NVlabs/ODISE.git
cd ODISE
PYTHONPATH=/path/to/repo:$PYTHONPATH python tools/train_net.py --config-file ${CONFIG_FILE} --num-gpus ${NUM_GPUS}
We consider ODISE as a code project, and would not compile and link it to site-packages. Therefore, the following example codes only ensures that you can run ODISE inside the folder.
git clone [email protected]:ViTAE-Transformer/DeepSolo.git
cd DeepSolo
python setup.py build_ext --inplace
PYTHONPATH=/path/to/repo:$PYTHONPATH python tools/train_net.py --config-file ${CONFIG_FILE} --num-gpus ${NUM_GPUS}
Please click the heading and follow our forked branch, the difference is:
- Adapt to the newest version of langchain (v0.0.225)
- Allow socks5 proxy to reach openai
- Validated at 394ffa7
We only demonstrate the easiest way to launch webui.
Note that we must replace REQS_FILE so that it would not automatically update our environment.
export REQS_FILE=/PATH-TO-Awesome-AI-Environment/requirements.txt
python launch.py --listen --port 4001 --disable-safe-unpickle --share