YOLO v3 Object Detector implemented in PyTorch

This repository contains PyTorch implementation of YOLOv3: An Incremental Improvement paper. Why this repo? The code implements the paper in pythonic/"pytorchish" style and will be familiar and (I hope) helpful for PyTorch users for understanding the paper/using YOLOv3 object detector. Original pretrained weights (on COCO dataset) were converted into PyTorch's state_dict and can be downloaded here (no need for complicated network builders and weights converters just common torch.load()).

Requirements

• Python 3
• PyTorch (0.4)
• torchvision
• OpenCV

Usage

The repo contains 2 runable scripts.

• Detection:

$ python detect.py images/dog.jpg

If you want to specify your own weights file use -weights path/to/weights or if you want to save the detection result use -save flag (Note: by default weights should be placed in data folder).

• Webcam Detection:

$ python webcam.py

Arguments are the same, except webcam.py doesn't have -save flag.

Examples

• 
• 
•