Vehicle Detection -- UMICH ROB 599 Final Project

Dependencies:

Windows opencv numpy darknet

Method:

Please refer to the PDF file for details.

Procedure to reproduce our result:

• 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.

Preview of the prediction of our network:

Our result:

On the kaggle competetion of the University of Michigan ROB 599 final project, we ranked 6 in the final.