ADSFormer

An Adaptive Dual Selective Transformer for Temporal Action Localization

Our code is built upon the codebase from ActionFormer,TemporalMaxer and Tridet, and we would like to express our gratitude for their outstanding work.

Environment

• Ubuntu20.04
• NVIDIA RTX A6000 GPU
• Python3.8, Pytorch2.0 and CUDA11.8
• h5py, joblib, matplotlib, numpy, pandas, PyYAML, scikit_learn, scipy, setuptools

Install NMS

cd ./libs/utils
python setup.py install --user
cd ../..

Pretrained Models

We provide our pre-trained models.

Baidu link：https://pan.baidu.com/s/1BLairYF4rJ-5fEYLWN3zgA?pwd=rnid password：rnid

Google drive：https://drive.google.com/drive/folders/1giaQytYcsHrVtFkyOJQRbO6mCulJEEVr?usp=sharing

Training and Evaluation

Train: python ./train.py ./configs/xxxx.yaml --save_ckpt_dir ./ckpt/xxx

Eval: python ./eval.py ./configs/xxxx.yaml <path of the weights>

Results