Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/LTContext/LTContext
[ICCV 2023] How Much Temporal Long-Term Context is Needed for Action Segmentation?
https://github.com/LTContext/LTContext
computer-vision deep-learning iccv2023 video-understanding
Last synced: 21 days ago
JSON representation
[ICCV 2023] How Much Temporal Long-Term Context is Needed for Action Segmentation?
- Host: GitHub
- URL: https://github.com/LTContext/LTContext
- Owner: LTContext
- License: other
- Created: 2023-08-16T13:47:46.000Z (about 1 year ago)
- Default Branch: main
- Last Pushed: 2024-06-21T09:52:25.000Z (5 months ago)
- Last Synced: 2024-07-05T13:54:32.644Z (4 months ago)
- Topics: computer-vision, deep-learning, iccv2023, video-understanding
- Language: Python
- Homepage: https://arxiv.org/abs/2308.11358
- Size: 334 KB
- Stars: 37
- Watchers: 3
- Forks: 3
- Open Issues: 2
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# LTContext
LTContext is an approach for temporal action segmentation, where it leverages
sparse attention to capture the long-term context of a video
and windowed attention to model the local information in the neighboring frames.
Here is an overview of the architecture:
The attention mechanism consist of Windowed Attention and Long-term Context Attention:
## Citation
If you use this code or our model, please cite our [paper](https://arxiv.org/abs/2308.11358):
```latex
@inproceedings{ltc2023bahrami,
author = {Emad Bahrami and Gianpiero Francesca and Juergen Gall},
title = {How Much Temporal Long-Term Context is Needed for Action Segmentation?},
booktitle = {IEEE International Conference on Computer Vision (ICCV)},
year = {2023}
}
```
## Installation
To create the [conda](https://docs.conda.io/en/latest/) environment run the following command:
```bash
conda env create --name ltc --file environment.yml
source activate ltc
```
## Dataset Preparation
The features and annotations of the Breakfast dataset can be downloaded from
[link 1](https://mega.nz/#!O6wXlSTS!wcEoDT4Ctq5HRq_hV-aWeVF1_JB3cacQBQqOLjCIbc8)
or
[link 2](https://zenodo.org/record/3625992#.Xiv9jGhKhPY).
### Assembly101
Follow the instructions at [Assembly101-Download-Scripts](https://github.com/assembly-101/assembly101-download-scripts) to download the TSM features.
The annotations for action segmentation can be downloaded from [assembly101-annotations](https://drive.google.com/drive/folders/1QoT-hIiKUrSHMxYBKHvWpW9Z9aCznJB7).
After downloading the annotation put `coarse-annotations` inside `data/assembly101` folder.
We noticed loading from `numpy` can be faster, you can convert the `.lmdb` features to `numpy` and use `LTContext_Numpy.yaml` config.
## Training and Evaluating the model
Here is an example of the command to train the model.
```bash
python run_net.py \
--cfg configs/Breakfast/LTContext.yaml \
DATA.PATH_TO_DATA_DIR [path_to_your_dataset] \
OUTPUT_DIR [path_to_logging_dir]
```
For more options look at `ltc/config/defaults.py`.
The value of `DATA.PATH_TO_DATA_DIR` for assembly101 should be the path to the folder containing the TSM features.
If you want to evaluate a pretrained model use the following command.
```bash
python run_net.py \
--cfg configs/Breakfast/LTContext.yaml \
DATA.PATH_TO_DATA_DIR [path_to_your_dataset] \
TRAIN.ENABLE False \
TEST.ENABLE True \
TEST.DATASET 'breakfast' \
TEST.CHECKPOINT_PATH [path_to_trained_model] \
TEST.SAVE_RESULT_PATH [path_to_save_result]
```
Check the [model card](MODEL_CARD.md) to download the pretrained models.
### Acknowledgement
The structure of the code is inspired by [SlowFast](https://github.com/facebookresearch/SlowFast).
The MultiHeadAttention is based on [xFormers](https://github.com/facebookresearch/xformers).
We thank the authors of these codebases.
## License
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.