Welcome to the EventSleep code repository!
| ArmsShake | Hands2Head | Head | LegsShake |
|---|---|---|---|
 |
 |
 |
 |
Here you will find a collection of useful scripts and tools designed to work with the EventSleep dataset.
To ensure a smooth setup, we have included a EventSleepEnv.yml file that lists all the necessary libraries and their versions. You can easily create a Python environment with the required dependencies, executing:
conda env create -f EventSleepEnv.ymlFurthermore, we have included a folder named "Toy_Data", which contains a small subset of the original data. This subset enables users to run the scripts without the need to download the entire dataset. In this way, this folder contains some isolated clips from a few trials. Most of the scripts include an option to specify if you want to run it either with the Toy_Data folder or with the original Dataset. Please note that to obtain the expected and accurate results, it is necessary to have access to the original complete dataset which is located in Synapse.
Additionally, we include a subfolder named "Models" which contains the trained models for our main approaches. It also serves as a repository for storing any new model trained using the scripts.
Here is a list of the scripts included in this folder:
-
data_tools.py: A script with a wide range of preprocessing and post-processing tools to work with EventSleep dataset (extract labels from folder names, labels dictionaries to obtain labels names, resize frames, crop frames to focus on the bed area, etc...).
-
events_to_frames.py: A script to enable the transformation of the event data into frames according the events frame based representation explained in the paper. As result, it generates a folder named "EventFrames" in which you will find the resulting frames saved in .npy format.
-
render_recordings.py: A script for visualizing the content of the event data, leveraging the frames representation and providing a synchronized view with the infrared recordings and the fine-grained ground truth labels. It generates a video that is located in Renders folder.
-
train_ResNet-E: A script to train ResNet-E model. As result, it generates a folder with the date as name that contains a checkpoint.pth with the snapshot of the trained model and a train_details.json file with the details of the run. This folder will be saved at Models/Events in the folder corresponding to the labels and configurations used to train.
-
train_ResNet-IR.py: A script to train ResNet-IR model. As result, it generates a folder with the date as name that contains a checkpoint.pth with the snapshot of the trained model and a train_details.json file with the details of the run. This folder will be saved at Models/Infrared in the folder corresponding to the labels and configurations used to train.
-
test_ResNet-E.py: A script to test a trained ResNet-E model model. As input, you must provide the checkpoint path stored in Models folder. As result, it prints and save in the checkpoint parent folder the confussion matrixes per configuration and averaged. Additionally, it saves a test_details.json file with the details of the run.
-
test_ResNet-IR.py: A script to test a trained ResNet-IR model from infrared frames. As input, you must provide the checkpoint path stored in Models folder. As result, it prints and save in the checkpoint parent folder the confussion matrixes per configuration and averaged. Additionally, it saves a test_details.json file with the details of the run.
We hope you find this repository helpful for your research and exploration of the EventSleep dataset.
If you find this repository useful, please cite our paper:
@article{plou2024eventsleep,
title={EventSleep: Sleep Activity Recognition with Event Cameras},
author={Plou, Carlos and Gallego, Nerea and Sabater, Alberto and Montijano, Eduardo and Urcola, Pablo and Montesano, Luis and Martinez-Cantin, Ruben and Murillo, Ana C},
journal={arXiv preprint arXiv:2404.01801},
year={2024}
}