Coding-Challenge-Sewts
Run :
- Dataset generation
- python3 dataset_generate.py OR python3 dataset_generate.py -p Location of dataset_raw
- Model Training
- python3 train.py --img 224 --batch 4 --epochs 2 --data /home/chandandeep/GitHub/Coding-Challenge-Sewts/Model/bcc.yaml --cfg /home/chandandeep/GitHub/Coding-Challenge-Sewts/Model/yolov5s.yaml --name BCCM
- tensorboard --logdir=runs
- Model Inference
- python yolov5/detect.py --source ../Coding-Challenge-Sewts/Dataset/Dataset_yolo/images/val --weights '../yolov5/runs/train/BCCM18/weights/best.pt'
- Output of the inference will be saved inside : ../yolov5/runs/train/BCCM
Goal :
Object (towel) localization (Bounding Box Detection)
Given :
Dataset of
- 100 images with towel
- Bounding box labels for each image in CSV
Steps :
- Load dataset images (JPEG) and BB_labels (TXT) files✓
- Analyse and display dataset✓
- Model selection (Yolo v5)✓
- Convert dataset to a model compatible format✓
- Train-test split✓
- Data pre-processing✓
- Model training✓
- Inference✓
Package Installation requirements :
Torch -
pip install torch==1.7.0+cpu torchvision==0.8.1+cpu torchaudio==0.7.0 -f https://download.pytorch.org/whl/torch_stable.html
Opencv -
pip3 install opencv-python
Yolo v5 -
git clone 'https://github.com/ultralytics/yolov5.git'
Pandas -
pip3 install pandas
YAML -
pip3 install pyyaml
Tensorboard -
pip3 install tensorboard
Matplotlib -
pip3 install matplotlib
Challenge Evaluation criteria :
• Sophisticated naming conventions✓
• Code styling (i.e. PEP8)✓
• Simple but expressive comments✓
• Code reusability✓
• Testing procedures✓
• Consistent use of a version control system (i.e. git)✓