anEEStudent / yolov5-Brainhack

Yolov5 CV model

These code bunches are what we've used for 2022 CDDC brainhack, finishing 5th in the finals. This is based on the Yolov5 model produced on github.

Since the code was originally written on colab, the ipynb is included for ease of access. However, we've removed all our training data.

Potential Training improvements

A higher contrasted photo increases ease of feature extraction, perhaps consider our contrast.py script to improve photo contrast

Training the model

python train.py --img 1280 --batch -1 --epochs 30 --data dataset.yaml --weights yolov5m.pt

Tuning the model

python train.py --img 640 --batch 16 --epochs 5 --data ./data/dataset.yaml --weights ./runs/train/exp40/weights/last.pt --evolve 10

Inference

Refer to our test_inference script

Download our weight

https://drive.google.com/file/d/1K_o3MmcvbketUv-8jys6YQZwz1pR-rSR/view?usp=sharing

Add the weight to the directory yolov5/runs/train/exp40/weights/

---

Collaborators:

@anEEStudent @GordonShinozaki @dominic-soh