Source Code for Machine Perception Course Project at ETH (263-3710-00L)

Dynamic Gesture Recognition Project Skeleton Code

Visit here for more information about the Machine Perception course.

All questions should first be directed to our course Piazza before being sent to my e-mail address.

Setup

The following two steps will prepare your environment to begin training and evaluating models.

1. Install dependencies by running (with sudo appended if necessary)

python3 setup.py install

2. Download training, validation and test datasets from kaggle project page.
3. Update data and output paths in config.py.
4. Train the model provided with source code by running

python3 training.py

4. When your model has completed training, it will perform a full evaluation on the test set. This output can be found in the folder runs/<experiment_name>/ as submission_<experiment_name>.csv.

Submit this csv file to our page on Kaggle.

Remarks

You can use external libraries. Make sure that you add them into setup.py before submitting your code.

Things to know

1. Branch master is the final project code branch which you can run for testing.
2. Branch source_code is the original code branch.
3. Branch working_project is the latest working branch where we have tested many things, such as C3D, batch normalization, weighted logit, data augmentation, bidirectional RNN, convolution feature maps drop out, moving average, etc.
4. Hyper-paramater configuration is included in the code.
5. If you want to run the script, follow the setup above directly!