# RelViT: Concept-guided Vision Transformer for Visual Relational Reasoning

Source: [RelViT: Concept-guided Vision Transformer for Visual Relational Reasoning](<https://www.sota2.com/research/paper/68fa6cc21e209131cea35d61>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-01-04T03:38:00.385Z

Authors: Xiaojian Ma, Weili Nie, Zhiding Yu, Huaizu Jiang, Chaowei Xiao, Yuke Zhu, Song-Chun Zhu, Anima Anandkumar

Published: 2022-04-24T02:46:43.000Z

## Abstract

Reasoning about visual relationships is central to how humans interpret the visual world. This task remains challenging for current deep learning algorithms since it requires addressing three key technical problems jointly: 1) identifying object entities and their properties, 2) inferring semantic relations between pairs of entities, and 3) generalizing to novel object-relation combinations, i.e., systematic generalization. In this work, we use vision transformers (ViTs) as our base model for visual reasoning and make better use of concepts defined as object entities and their relations to improve the reasoning ability of ViTs. Specifically, we introduce a novel concept-feature dictionary to allow flexible image feature retrieval at training time with concept keys. This dictionary enables two new concept-guided auxiliary tasks: 1) a global task for promoting relational reasoning, and 2) a local task for facilitating semantic object-centric correspondence learning. To examine the systematic generalization of visual reasoning models, we introduce systematic splits for the standard HICO and GQA benchmarks. We show the resulting model, Concept-guided Vision Transformer (or RelViT for short) significantly outperforms prior approaches on HICO and GQA by 16% and 13% in the original split, and by 43% and 18% in the systematic split. Our ablation analyses also reveal our model's compatibility with multiple ViT variants and robustness to hyper-parameters.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Human-Object Interaction Recognition on HICO (Systematic-easy)](<https://www.sota2.com/research/sota/human-object-interaction-recognition-on-hico-systematic-easy.md>): Human-Object Interaction Recognition; HICO (Systematic-easy); mAP (Full): 37.21.

- [Human-Object Interaction Recognition on HICO (Systematic-hard)](<https://www.sota2.com/research/sota/human-object-interaction-recognition-on-hico-systematic-hard.md>): Human-Object Interaction Recognition; HICO (Systematic-hard); mAP (Full Classes): 23.06.

- [Visual Question Answering on GQA (Orig.)](<https://www.sota2.com/research/sota/visual-question-answering-on-gqa-orig.md>): Visual Question Answering; GQA (Orig.); Overall Accuracy: 57.87.

- [Visual Question Answering on GQA (Sys.)](<https://www.sota2.com/research/sota/visual-question-answering-on-gqa-sys.md>): Visual Question Answering; GQA (Sys.); Overall Accuracy: 35.48.

- [Human-Object Interaction Detection on HICO-DET (test)](<https://www.sota2.com/research/sota/human-object-interaction-detection-on-hico-det-test.md>): Human-Object Interaction Detection; HICO-DET (test); mAP (full): not recorded.

- [Human-Object Interaction Detection on HICO-DET](<https://www.sota2.com/research/sota/human-object-interaction-detection-on-hico-det.md>): Human-Object Interaction Detection; HICO-DET; mAP (Full): not recorded.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2204.11167>)

[PDF](<https://arxiv.org/pdf/2204.11167.pdf>)