claydon-wang / oftta Goto Github PK

This is the official implementation for "Optimization-Free Test-Time Adaptation for Cross-Person Activity Recognition".

Please create and activate the following conda envrionment.

# It may take several minutes for conda to solve the environment
conda create -y -n oftta python=3.9
conda activate oftta
pip install -r requirements.txt 

Three datasets (UCI-HAR, Opportunity, and UniMiB-SHAR) are utilized in the experiments. The pre-processed outcome can be downloaded from here. The datasets is adopt from GILE. Please save datasets under folder ./data.

Since test-time adaptation needs pre-trained on source domains. We provide our used model in Table 4. You can download the model from here. Please save datasets under folder ./ckpt. If you want to train your model from scratch, you can refer to code for generalizable HAR.

To reproduce the leave-one-out adaptation results in Table 4, you just need:

bash adapt.sh

You can get the all results in Table 4.

We support all the TTA algorithms used in the paper. Feel free to adopt them on other types of dataset.