github.com/giopaglia/interactionwise-vrd

Interactionwise would like to be a reference framework for Visual Relationship Detection & Scene Graph Generation.

Open this visualization on its own page →

Contributors

10

Lines of Code

7,527

From

2018-01-29

To

2020-06-02

About giopaglia/interactionwise-vrd

Interactionwise is a research framework focused on Visual Relationship Detection and Scene Graph Generation, areas of computer vision that identify relationships between objects in images. The project is built on prior work in structural ranking for visual relationship detection and provides code from a thesis effort aimed at understanding semantic connections within images.

The repository includes infrastructure for handling multiple datasets, particularly the VRD (Visual Relationship Detection) and Visual Genome datasets, along with pre-trained models including VGG16 for feature extraction, Word2vec embeddings, and Faster R-CNN for object detection. The codebase requires various Python packages and pre-downloaded assets to function, though the authors acknowledge the project is currently in a messy state with setup instructions still being organized.

This is a specialized academic resource intended for researchers working on scene understanding and relationship detection in computer vision. The project serves as a reference implementation for semantic awareness in visual relationship tasks, building directly on established detection frameworks rather than creating entirely novel architectures.

Share this video