Image scene graph generation benchmark
Scene Graph Benchmark in PyTorch 1.4
This project aims at providing the necessary building blocks for easily creating detection and segmentation models using PyTorch 1.0.
Highlights
- Upgrad to pytorch 1.4 (can also upgrade to 1.7)
- Multi-GPU training and inference
- Batched inference: can perform inference using multiple images per batch per GPU.
- Fast and flexible tsv dataset format
- Remove FasterRCNN detector dependency: during relation head training, can plugin bounding boxes from any detector.
- Provides pre-trained models for different scene graph detection algorithms (IMP, MSDN, GRCNN, Neural Motif, RelDN).
- Provides bounding box level and relation level feature extraction functionalities
- Provides large detector backbones (ResNxt152)
Installation
Check INSTALL.md for installation instructions.
Model Zoo and Baselines
Pre-trained models can be found in SCENE_GRAPH_MODEL_ZOO.md