This repository is final-project of CCU_DL. Given single image or video, the goal is detect and recognize category of merchandise on shelf.
git clone --recurse-submodules https://github.com/Yenhongxuan/DL-Final-Project.git
If you have already cloned and forgot to use --recurse-submodules, run git submodule updata --init
Step 1.
conda create -n env_name python=3
Step 2.
pip3 install -r requirements.txt
Link below is dataset we used. We collect these image by downloading from internet and taking photo at convenience store.
https://drive.google.com/drive/folders/1HWXsozNFpdiMKPjGbrHirxvPiqkWc50o?usp=sharing
Our dataset is split to six categories, which are beverage, bread, cookie, frozen-food, riceball, instant-noodles
You can excute like the following to train YOLOv5
cd yolov5/
python train.py --data SKU-110K.yaml --weights yolov5n.pt --img 640
The weight will be stored in ./yolov5/runs/train/exp/weights/
Options for train.py are as following
--root Root directory of data
--epochs Number of epoch to train
--lr Learning rate
--bs Batchsize
--balance_sample Whether apply weighted sampleing during training
python3 train.py
You can excute the following to train Grocery recognition network
python3 train.py --root 'data_fir' --balance_sample
Options for inference.py are as following:
--source Path of image or video to be inferenced
--weights Weight of YOLO to be used
--best_model Path of Check-point of ResNet model
--num_classes Amount of category to be predicted
You can excute like this
python3 inference.py --source ./sample_img/1.jpg --weights ./yolo_weights/yolo_best.pt --best_model ./ResNet_weight/best_model_1.pt