Skip to content

Latest commit

 

History

History
35 lines (20 loc) · 1.26 KB

run.md

File metadata and controls

35 lines (20 loc) · 1.26 KB
Raw

Multi-scale Conditional Relationship Graph Network for Referring Relationships in Images

This is the source code for our paper [Multi-scale Conditional Relationship Graph Network for Referring Relationships in Images].

Environment (Recommended)

  • Ubuntu 16.04
  • CUDA 10.0
  • python 3.6
  • pytorch 1.3

Data preparation

Download urls

process commands for vrd dataset

python data.py --save-dir data/dataset-vrd --img-dir data/vrd_img/sg_test_images/ --test --image-metadata data/VRD/test_image_metadata.json --annotations data/VRD/annotations_test.json --save-images python data.py --save-dir data/dataset-vrd --img-dir data/vrd_img/sg_train_images/ --image-metadata data/VRD/train_image_metadata.json --annotations data/VRD/annotations_train.json --save-images

Training, validation and test

python train_vrd.py

Citation

If you are using this repository, please use the following citation: Jian Zhu and Hanli Wang, Multi-scale Conditional Relationship Graph Network for Referring Relationships in Images, IEEE Transactions on Cognitive and Developmental Systems, 2021.