8000 GitHub - xyh97/UNICORN: Source code for UNICORN, a unified causal video-oriented language modeling framework.
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

xyh97/UNICORN

UNICORN

Installation

  1. Creating conda environment
conda create -n unicorn python=3.10
conda activate unicorn
  1. Build from source
cd UNICORN
pip install -r requirements.txt
pip install -e .
  1. Unzip data & Train

download the folder /nfs/data/data/howto100m/annotations to your server in howto100m folder.

unzip activitynet.zip -d ./activitynet
unzip youcook2.zip -d ./youcook2
unzip qvhighlights.zip -d ./qvhighlights
unzip Charades_v1_480.zip -d ./charades

cd youcook2
mv video_1fps video_1fps_new
bash run_scripts/unicorn/train/pretrain_75frames_temp_aug.sh

Acknowledgement

LAVIS from Salesforce: the codebase we built upon, and we selected InstructBLIP as our base model initialization.

About

Source code for UNICORN, a unified causal video-oriented language modeling framework.

Resources

License

Code of conduct

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published
0