Transformer-based Fall Detection in Videos

Configuration files

There are two configuration files: config_eval1.yaml and config_eval2.yaml, used for the evaluation strategy 1 and the evaluation strategy 2, respectively. Within them there are several options to tune the experiments. The values in the repository were the ones used to obtain the best results.

Data

The code expects the UP-Fall dataset format, i.e.

├── Subject 1
│   ├── Activity1
|   |   ├── Trial1
|   |   |   ├── Camera1
|   |   |   |    ├── frame_0000.png
|   |   |   |    ├── frame_0001.png
|   |   ├── Trial 2
|   |   └── Trial 3
│   └── Activity2
├── Subject 2

The folder containing this tree structure is declared in the configuration file using the option dataset_folder.

Running the code

Modify main.py to include the desired configuration file (config_eval1.yaml by default) in line 21. Then call:

python main.py

Citation

If you plan to use this code, please, cite us:

TODO