# Shuffle-Then-Assemble: Learning Object-Agnostic Visual Relationship Features

arXiv: https://arxiv.org/abs/1808.00171 · HTML: https://mlsanity.com/paper/1808.00171/

## Key facts

- Cited by 11 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 1 Aug 2018.

## Built on by (most cited first, 11 of 11)

- [A Comprehensive Survey of Scene Graphs: Generation and Application](https://mlsanity.com/paper/2104.01111/) (2021)
- [Scene Graph Generation: A Comprehensive Survey](https://mlsanity.com/paper/2201.00443/) (2022)
- [SINC: Spatial Composition of 3D Human Motions for Simultaneous Action Generation](https://mlsanity.com/paper/2304.10417/) (2023)
- [Fully Convolutional Scene Graph Generation](https://mlsanity.com/paper/2103.16083/) (2021)
- [REGRAD: A Large-Scale Relational Grasp Dataset for Safe and Object-Specific Robotic Grasping in Clutter](https://mlsanity.com/paper/2104.14118/) (2021)
- [ACP++: Action Co-occurrence Priors for Human-Object Interaction Detection](https://mlsanity.com/paper/2109.04047/) (2021)
- [Weakly Supervised Human-Object Interaction Detection in Video via Contrastive Spatiotemporal Regions](https://mlsanity.com/paper/2110.03562/) (2021)
- [Neural Message Passing for Visual Relationship Detection](https://mlsanity.com/paper/2208.04165/) (2022)
- [Tackling the Challenges in Scene Graph Generation with Local-to-Global Interactions](https://mlsanity.com/paper/2106.08543/) (2021)
- [Semi-Supervised Image Captioning by Adversarially Propagating Labeled Data](https://mlsanity.com/paper/2301.11174/) (2023)
- [Towards Unseen Triples: Effective Text-Image-joint Learning for Scene Graph Generation](https://mlsanity.com/paper/2306.13420/) (2023)

## Abstract (arXiv)

Due to the fact that it is prohibitively expensive to completely annotate visual relationships, i.e., the (obj1, rel, obj2) triplets, relationship models are inevitably biased to object classes of limited pairwise patterns, leading to poor generalization to rare or unseen object combinations. Therefore, we are interested in learning object-agnostic visual features for more generalizable relationship models. By "agnostic", we mean that the feature is less likely biased to the classes of paired objects. To alleviate the bias, we propose a novel textttShuffle-Then-Assemble pre-training strategy. First, we discard all the triplet relationship annotations in an image, leaving two unpaired object domains without obj1-obj2 alignment. Then, our feature learning is to recover possible obj1-obj2 pairs. In particular, we design a cycle of residual transformations between the two domains, to capture shared but not object-specific visual patterns. Extensive experiments on two visual relationship benchmarks show that by using our pre-trained features, naive relationship models can be consistently improved and even outperform other state-of-the-art relationship models. Code has been made available at: urlhttps://github.com/yangxuntu/vrd.