rorubyy/thermal_rgb_fusion_yolov8

Set up environment

1. go into workspace yolov8

   cd yolov8

2. create docker image

   cd docker

   docker build -t ultralytics .

3. set up docker environment

   cd ..

   docker compose up -d --build lab

4. goto container workspace -> /root/code
5. install requirements

   pip install -r requirements.txt 

Train

1. preprare thermal + RGB pair data
2. create yaml file and set your custom dataset path in /root/code/ultralytics/yolo/cfg/xxx.yaml
3. your custom dataset should put like this

   train_rgb: /root/code/ultralytics/yolo/datasets/LLVIP/RGB/images/train
   val_rgb: /root/code/ultralytics/yolo/datasets/LLVIP/RGB/images/val
   train_ir: /root/code/ultralytics/yolo/datasets/LLVIP/IR/images/train
   val_ir: /root/code/ultralytics/yolo/datasets/LLVIP/IR/images/val
   test_rgb: /root/code/ultralytics/yolo/datasets/LLVIP/RGB/images/val
   test_ir: /root/code/ultralytics/yolo/datasets/LLVIP/IR/images/val

4. edit default.yaml

   data: xxx.yaml

5. train with your curstom dataset

   python train.py

Predict

1. preprare thermal + RGB pair data
2. edit default.yaml which in /root/code/ultralytics/yolo/cfg/

   • input data

     data type should be a photo pair or a directory with lot of pair photos

     source_rgb: "/root/code/ultralytics/yolo/datasets/LLVIP/RGB/images/val"
     source_ir: "/root/code/ultralytics/yolo/datasets/LLVIP/IR/images/val"

   • model

     model: /root/code/best.pt # path to model file, i.e. yolov8n.pt, yolov8n.yaml

3. predict

   python predict_twostream.py