Final project of Wang Shuai

Based on a Pytorch implementation of YOLOv3, prompt three improvement for YOLOv3.
If any problems met, please e-mail to [email protected] with no hesitation.

Prepare work

Clone and install requirements

$ git clone https://github.com/Dan944/project

Set the environment

$ pip install -r requirements.txt

Strongly recommand setting CUDA and CUDNN for GPU training

Running on COCO

Result show

Download pretrained weights

$ cd weights/
$ bash download_weights.sh

Detect the image in samples

$ python3 detect.py

Download COCO dataset

$ cd data/
$ bash get_coco_dataset.sh

Evaluates the model on COCO test.

$ python test.py --weights_path weights/yolov3.weights

For further training

$ python train.py --model_def config/yolov3.cfg --data_conig config/coco.data

Run on PASCAL VOC (or other dataset)

Prepare

1.Download the data.
2.Save the images in project/data/custom/images, save the labels in project/data/custom/labels.
3.Save the paths of train data in project/data/custom/train.txt, save the paths of test data in project/data/custom/valid.txt.
A sample file project/data/custom/transform_voc_data.py could be referenced for step 2,3
(If select VOC, step 1-3 is enough) 4.run the code config/create_custom_model.sh <num-classes>
5.Change class names to data/custom/classes.names

Train

$ python train.py --model config/yolov3-custom.cfg --data config/custom.data

Add --pretrained_weights weights/darknet53.conv.74 to train using a backend pretrained on ImageNet.

Test

$ python test.py --model config/yolov3-custom.cfg --data config/custom.data --weight_path <your trained weight>

Credit

@article{yolov3,
  title={YOLOv3: An Incremental Improvement},
  author={Redmon, Joseph and Farhadi, Ali},
  journal = {arXiv},
  year={2018}
}