MattKleinsmith/void-detector

Detect voids in grocery store shelves.

# void-detector Detect voids in grocery store shelves. The above are predictions from the first model trained. The images are from the test set. ## Table of Contents - Demo - Visualization - Goal - Log - 2018-02-15: Collect and label data - 2018-02-16: Train model - 2018-02-17: Inspect and report results - 2018-02-18: Train with voidless images - 2018-02-19: Understand the performance drop - 2018-02-20: Clean the code - 2018-02-21: Log results in a database (part 1) - 2018-02-22: (Take a break) - 2018-02-23: Log results in a database (part 2) - 2018-02-24: Log results in a database (part 3) - Ambitions ## Demo The demo will draw bounding boxes on copies of the given images, similar to the GIF above. The results will be stored in `IMAGE_DIR/void-detector-outputs/`. ### Dockerless `python demo.py IMAGE_DIR --dockerless` Requirements: - [Anaconda 5, Python 3.6 version](https://www.anaconda.com/download/#linux) - pip install opencv-python - conda install pytorch torchvision cuda90 -c pytorch - pip install tqdm ### Docker `python demo.py IMAGE_DIR` Requirements: - [nvidia-docker 2.0](https://github.com/nvidia/nvidia-docker/wiki/Installation-(version-2.0)) - Access to the Docker image - Option 1: Obtain read permissions for the Docker image at matthewkleinsmith/void-detector - Option 2: Build the image with `docker build -t matthewkleinsmith/void-detector docker/.` # Visualization # Goal - Input: Image of shelf - Output: Bounding boxes around voids # Log ## 2018-02-15: Collect and label data - [x] Collect images of grocery store shelves - I walked down the aisles of my local grocery store while recording the shelves - - I used a grocery cart to stable my smartphone's camera and to bring the height of the camera close to t