LINKIT Intermediate Challenge (WS22-23) Real Time Hand Gesture Recognition

Linkit WS22-23- Intermediate Challenge

Setup

1. Install Dependencies

pip install -r requirements.txt Note: Requires Python 3.6-3.10

2. Open challenge.ipynb

Repository Structure

.
├── README.md
├── challenge.ipynb
├── datasets
│   ├── yourdatasetname 
│   │   ├── images
│   │   │   ├── train
│   │   │   │   ├── 0.jpg
│   │   │   │   ├── 1.jpg
│   │   │   ├── val 
│   │   │   ├── test-dev
│   │   ├── labels
│   │   │   ├── train
│   │   │   │   ├── 0.txt
│   │   │   │   ├── 1.txt
│   │   │   ├── val 
│   │   │   ├── test-dev
│── .gitignore
├── requirements.txt
├── yolov5 [GIT SUBMODULE]

Training

Monitoring

It is important to monitor the training process to ensure that the model is training properly. To do so, we recommend Weights and Biases (or tensorboard). Both tools keep track of the training process and automatically log the results.

Weights and Biases

1. Create an account on Weights and Biases
2. Install the wandb package pip install wandb
3. Login to your account wandb login
4. Run the training script with the --project flag python train.py --project <project_name>
5. Go to your Weights and Biases dashboard to view the results

For more information on the YOLOV5 integration with Weights and Biases, refer to here

Tensorboard

If installed, training will automatically log to tensorboard.

Submission

1. Fill in the inference function in challenge.ipynb

   1. Inference expects a model
   2. Note: Feel free to borrow code from the yolov5 detect.py

2. Update your requirements.txt file if you have installed any additional packages: pip list --format=freeze > requirements.txt

3. We will run the code with github actions using the following command: python submission.py

4. The Output of the code should be a pandas dataframe file in the same format as the exemplary inference functions output

5. You can check your current fps via the actions tab on github

   1. Open the actions tab
   2. Click on your latest action
   3. Open the Run Challenge step and scroll down to the Test Results section

Evaluation

We will evaluate your model on the following metrics:

1. Mean Average Precision (mAP)
2. FPS (Frames per second) must be greater than 5
3. Model Size

BONUS: Can you convert and run your model in the onnx format? This may increase inference speed and reduce model size.

Contact

If you have any questions, please contact us on linkit-Teams (or email us) at

• Philipp Kiesling: philipp.kiesling@linkit.tech
• Justus Thomsen: justus.thomsen@linkit.tech