Divide-and-Conquer Predictor for Unbiased Scene Graph Generation

Implement of paper Divide-and-Conquer Predictor for Unbiased Scene Graph Generation.

Contents

Overview

Installation

Dataset

Training

Pre-trained DCNet

Evaluation

Overview

We divide the predicate prediction into a few sub-tasks with a Divide-and-Conquer Predictor (DC-Predictor).

Installation

Check INSTALL.md for installation instructions.

Dataset

Check DATASET.md for instructions of dataset preprocessing.

Training

Prepare Faster-RCNN Detector

We adopted the pretrained Faster R-CNN provided by Scene Graph Benchmark. You can download the pretrained Faster R-CNN and put it into the folder:

/home/username/checkpoints/pretrained_faster_rcnn

Scene Graph Generation Model

This code include DCNet and other SGG methods.

There are three standard protocols: 1) Predicate Classification (PredCls). 2) Scene Graph Classification (SGCls), and 3) Scene Graph Detection (SGDet).

We use MODEL.ROI_RELATION_HEAD.USE_GT_BOX and MODEL.ROI_RELATION_HEAD.USE_GT_OBJECT_LABEL to select the protocols.

Test Example 1: (Neural Motifs, SGCls)

Test Example 2: (DCNet, PredCls)

Test Example 3: (DCNet, SGDet)

Checkpoint DCNet

The checkpoint of are provided in this link.

Evaluation

Test Example: (SGCls)

Citations

If you find this project helps your research, please kindly consider citing our papers in your publications.

Acknowledgment

This repository is developed on top of the scene graph benchmarking framwork develped by KaihuaTang