-
Download the challenge data to
/data/behavior-representation/
following the instructions at https://sites.google.com/view/mabe22/home.- Extract the submission video data to
/data/behavior-representation/videos/full_size/submission/
.
- Extract the submission video data to
-
Build a docker image using the Dockerfile:
docker build . -t mabe_2022
. -
Enter the docker container with
docker run -it --gpus '"device=1"' --shm-size 2g -v SRC_DIR/mabe_2022:/app -v /data/behavior-representation/:/data/behavior-representation -e PYTHONPATH=/app -w /app mabe_2022:latest bash
. -
Embed all frames with BEiT:
python3 utils/embed_frames_beit.py
. This takes ~2.5 days on an Nvidia RTX 3080 GPU. -
Run
utils/average_motion.py
. This computes a measure of the amount of motion in each frame, based on the keypoints. -
Run
utils/train_simclr_model.py
to train a SimCLR model and use it to compute an embedding for each frame. -
Run
utils/handcrafted_geometries.py
to compute a bunch of handcrafted features based on the keypoints. -
Run
combine_embeddings.ipynb
to combine BEiT embeddings, SimCLR embeddings and handcrafted features with a weighted PCA transform. -
Run
append_mean_beit_pca.ipynb
to exchange the last 8 dimensions of the previous PCA with the first 8 PCA components of BEiT, averaged over each video snippet.
-
Notifications
You must be signed in to change notification settings - Fork 0
License
IRLAB-Therapeutics/mabe_2022
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
About
No description, website, or topics provided.
Resources
License
Stars
Watchers
Forks
Releases
No releases published
Packages 0
No packages published