Skip to content

Latest commit

 

History

History
49 lines (42 loc) · 1.26 KB

README.md

File metadata and controls

49 lines (42 loc) · 1.26 KB
Raw

Introduction

YOLO nano is from this paper.

TODO

Since I'm too busy at the end of the semester, I will continue working on this project after my exams.

  • Finish a draft version of implementation
  • Add README
  • Add checkpoint support
  • Add COCO dataset support (Code still needs cleaning. I'm working on it.)
  • Add multi scale and horizontal flip transforms
  • Reconstruct the code of visualizer
  • Add val and test
  • Add VOC support
  • Test accuracy

Installation

git clone https://github.com/liux0614/yolo_nano
pip3 install -r requirements.txt

COCO

Project Structure

root/
  results/
  datasets/
    coco/
      images/
        train/
        val/
      annotation/
        instances_train2017.json
        instances_val2017.json

Train

To use COCO dataset loader, pycocotools should be installed via the following command.

pip install "git+https://github.com/philferriere/cocoapi.git#egg=pycocotools&subdirectory=PythonAPI"

To train on COCO dataset:

python3 main.py --dataset_path datasets/coco/images --annotation_path datasets/coco/annotation/instances_train2017.json 
                --dataset coco --lr 0.0001 --conf_thres 0.8 --nms_thres 0.5