mlsanityData to 14 Oct 2026

Compositional Feature Augmentation for Unbiased Scene Graph Generation

arXiv 2308.06712 ↗First posted 13 Aug 2023cs.CVcited by 15

Not on the map. Why?

Abstract (arXiv)

Scene Graph Generation (SGG) aims to detect all the visual relation triplets <textttsub, textttpred, textttobj> in a given image. With the emergence of various advanced techniques for better utilizing both the intrinsic and extrinsic information in each relation triplet, SGG has achieved great progress over the recent years. However, due to the ubiquitous long-tailed predicate distributions, today's SGG models are still easily biased to the head predicates. Currently, the most prevalent debiasing solutions for SGG are re-balancing methods, eg, changing the distributions of original training samples. In this paper, we argue that all existing re-balancing strategies fail to increase the diversity of the relation triplet features of each predicate, which is critical for robust SGG. To this end, we propose a novel Compositional Feature Augmentation (CFA) strategy, which is the first unbiased SGG work to mitigate the bias issue from the perspective of increasing the diversity of triplet features. Specifically, we first decompose each relation triplet feature into two components: intrinsic feature and extrinsic feature, which correspond to the intrinsic characteristics and extrinsic contexts of a relation triplet, respectively. Then, we design two different feature augmentation modules to enrich the feature diversity of original relation triplets by replacing or mixing up either their intrinsic or extrinsic features from other samples. Due to its model-agnostic nature, CFA can be seamlessly incorporated into various SGG frameworks. Extensive ablations have shown that CFA achieves a new state-of-the-art performance on the trade-off between different metrics.

Key facts. Cited by 15 papers in the map (data through 14 Oct 2026). Builds on 45 of its 51 references that are in the map. First posted on arXiv 13 Aug 2023.

Builds on

45 of 51 references.

References · 45

  1. 2017Attention Is All You Needcited by 60,361
  2. 2020A Simple Framework for Contrastive Learning of Visual Representationscited by 12,049
  3. 2015Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networkscited by 9,639
  4. 2017mixup: Beyond Empirical Risk Minimizationcited by 5,183
  5. 2016Feature Pyramid Networks for Object Detectioncited by 5,036
  6. 2016Visual Genome: Connecting Language and Vision Using Crowdsourced Dense Image Annotationscited by 2,656
  7. 2017Improved Regularization of Convolutional Neural Networks with Cutoutcited by 1,628
  8. 2017Random Erasing Data Augmentationcited by 1,239
  9. 2019LVIS: A Dataset for Large Vocabulary Instance Segmentationcited by 1,217
  10. 2019Large-Scale Long-Tailed Recognition in an Open Worldcited by 705
  11. 2017Scene Graph Generation by Iterative Message Passingcited by 462
  12. 2017Neural Motifs: Scene Graph Parsing with Global Contextcited by 414
33 more
  1. 2020Unbiased Scene Graph Generation from Biased Trainingcited by 356
  2. 2016Visual Relationship Detection with Language Priorscited by 334
  3. 2018Learning to Compose Dynamic Tree Structures for Visual Contextscited by 259
  4. 2019Knowledge-Embedded Routing Network for Scene Graph Generationcited by 164
  5. 2021Bipartite Graph Network with Adaptive Message Passing for Unbiased Scene Graph Generationcited by 137
  6. 2020Feature Space Augmentation for Long-Tailed Datacited by 136
  7. 2020Visual Compositional Learning for Human-Object Interaction Detectioncited by 130
  8. 2017Visual Translation Embedding Network for Visual Relation Detectioncited by 112
  9. 2021Detecting Human-Object Interaction via Fabricated Compositional Learningcited by 79
  10. 2020PCPL: Predicate-Correlation Perception Learning for Unbiased Scene Graph Generationcited by 75
  11. 2021Recovering the Unbiased Scene Graphs from the Biased Onescited by 71
  12. 2022Stacked Hybrid-Attention and Group Collaborative Learning for Unbiased Scene Graph Generationcited by 67
  13. 2021Learning of Visual Relations: The Devil is in the Tailscited by 65
  14. 2020CogTree: Cognition Tree Loss for Unbiased Scene Graph Generationcited by 62
  15. 2022The Devil is in the Labels: Noisy Label Correction for Robust Scene Graph Generationcited by 60
  16. 2022Fine-Grained Scene Graph Generation with Data Transfercited by 57
  17. 2018Counterfactual Critic Multi-Agent Training for Scene Graph Generationcited by 53
  18. 2021From General to Specific: Informative Scene Graph Generation via Balance Adjustmentcited by 51
  19. 2023Visually-Prompted Language Model for Fine-Grained Scene Graph Generation in an Open Worldcited by 35
  20. 2022Fine-Grained Predicates Learning for Scene Graph Generationcited by 27
  21. 2023Zero-shot Visual Relation Detection via Composite Visual Cues from Large Language Modelscited by 27
  22. 2022Resistance Training using Prior Bias: toward Unbiased Scene Graph Generationcited by 22
  23. 2022NICEST: Noisy Label Correction and Training for Robust Scene Graph Generationcited by 19
  24. 2022Hierarchical Memory Learning for Fine-Grained Scene Graph Generationcited by 15
  25. 2020Generative Compositional Augmentations for Scene Graph Predictioncited by 14
  26. 2022Label Semantic Knowledge Distillation for Unbiased Scene Graph Generationcited by 14
  27. 2021Semantic Compositional Learning for Low-shot Scene Graph Generationcited by 12
  28. 2020HOSE-Net: Higher Order Structure Embedded Network for Scene Graph Generationcited by 9
  29. 2021Feature Generation for Long-tail Classificationcited by 9
  30. 2022Rethinking the Evaluation of Unbiased Scene Graph Generationcited by 9
  31. 2023Skew Class-balanced Re-weighting for Unbiased Scene Graph Generationcited by 5
  32. 2022Peer Learning for Unbiased Scene Graph Generationcited by 4
  33. 2022Unbiased Scene Graph Generation using Predicate Similaritiescited by 1

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2023: 1; 2024: 9; 2025: 3; 2026: 2.

15 papers.

Sort:

By area:

  1. 2024EGTR: Extracting Graph from Transformer for Scene Graph Generationcited by 47
  2. 2023LLM4SGG: Large Language Models for Weakly Supervised Scene Graph Generationcited by 30
  3. 2024Scene Graph Generation with Role-Playing Large Language Modelscited by 16
  4. 2025A Causal Adjustment Module for Debiasing Scene Graph Generationcited by 14
  5. 2024Neural Clustering based Visual Representation Learningcited by 12
  6. 2025Universal Scene Graph Generationcited by 11
  7. 2024Hydra-SGG: Hybrid Relation Assignment for One-stage Scene Graph Generationcited by 10
  8. 2024Leveraging Predicate and Triplet Learning for Scene Graph Generationcited by 8
  9. 2024Scene-Graph ViT: End-to-End Open-Vocabulary Visual Relationship Detectioncited by 6
  10. 2025Learning 4D Panoptic Scene Graph Generation from Rich 2D Visual Scenecited by 6
  11. 2024Scene Graph Generation Strategy with Co-occurrence Knowledge and Learnable Term Frequencycited by 4
  12. 2024Semantic Diversity-aware Prototype-based Learning for Unbiased Scene Graph Generationcited by 4
3 more
  1. 2024Fine-Grained Scene Graph Generation via Sample-Level Bias Predictioncited by 1
  2. 2026BUSSARD: Normalizing Flows for Bijective Universal Scene-Specific Anomalous Relationship Detectioncited by 0
  3. 2026Learning Context-Conditioned Predicate Semantics via Prototype Feedbackcited by 0