Windows opencv numpy darknet
Please refer to the PDF file for details.
- download the dataset from ROB599 Dataset and extract to /deploy
- run main.py to preprocess data
- run train/obj/pprocess.py to generate list for validation set
- download pretrained weight file to /train
- run train/train.bat to start training. This may take 5~8 hours
- follow instruction from darknet repo to decide which weight to use
- run train/obj_test.bat to generate raw results. You may need to modify the path to the weight file.
- run train/translate_yolo_finetune.py to generate final .csv file.
On the kaggle competetion of the University of Michigan ROB 599 final project, we ranked 6 in the final.