charmingwang / sirn

SIRN

SIRN is a network for Transmission Line dense-tiny object detection. This repo is the implementation of the paper ("SIRN: An Iterative Reasoning Network for Transmission Lines Based on Scene Prior Knowledge").

Requirements

• Python3.8
• Python packages
  • PyTorch >= 1.0
  • Torchvision >= 0.9.0
  • opencv-python-headless
  • fvcore
  • cloudpickle
  • omegaaconf
  • pycocotools
  • tidecv
  • fairscale
  • timm
  • scikit-learn

Get Started

After successfully completing requirements, you can be ready to run the demo.

• Download the test.pth which finally use in the paper(SIRN) from [Weights](链接：https://pan.baidu.com/s/1NSEzvPzPby5FNv6IwZKiJw (extract code:8qb9)

• Download the datasets from [Datasets](https://pan.baidu.com/s/1atcDoJ2pDaXF8uSZh_3kvg (extractcode:8lsq)

• Put datasets into the:

{repo_root}/

• Put test.pth into the:

{repo_root}/

• Using this code to see the detection results in the Transmission line datasets:

python train_net.py --config-file ./model_configs/faster_r50_s600.yaml --eval-only MODEL.WEIGHTS test.pth