Dice number recognition

Semestral task for Cognitive Systems course at CTU in Prague.

The goal is to implement a recognition tool for standard 6-sided dice. Ideally inspired by human brain function.

Runtime

Final implementation uses the well known YOLOv5 neural model. To run, firstly install the requirements.

$ pip install -r requirements.txt

Then, since I added the trained models to this repository for simplicity, just run the inference:

$ python yolov5/detect.py --weights best_models/best_medium.pt --source 0 --conf-thres 0.7

Then, you can see the full runtime in yolov5/runs/detect/exp*/0.mp4 folder, (* symbolizes a number, depending on how many detections you had)

To run inference on a directory of images (i.e. Dice-3/test/images/), run:

$ python yolov5/detect.py --weights best_models/best_medium.pt --source Dice-3/test/images/ --conf-thres 0.7

Then, you can see the inferred images in yolov5/runs/detect/exp*/ folder, (* symbolizes a number, depending on how many detections you had)

You can set up the inference confidence treshold using the --conf-thres parameter. Decide for the best

Full report about the work (in czech) is in the report.pdf file. A presentation of the work (also in czech) is in the presentation folder.