mlsanityData to 7 Oct 2026

Can CLIP Count Stars? An Empirical Study on Quantity Bias in CLIP

arXiv 2409.15035 ↗First posted 23 Sep 2024cs.CV cs.CLcited by 9

Not on the map. Why?

Abstract (arXiv)

CLIP has demonstrated great versatility in adapting to various downstream tasks, such as image editing and generation, visual question answering, and video understanding. However, CLIP-based applications often suffer from misunderstandings regarding user intent, leading to discrepancies between the required number of objects and the actual outputs in image generation tasks. In this work, we empirically investigate the quantity bias in CLIP. By carefully designing different experimental settings and datasets, we comprehensively evaluate CLIP's understanding of quantity from text, image, and cross-modal perspectives. Our experimental results reveal a quantity bias in CLIP embeddings, impacting the reliability of downstream tasks.

Key facts. Cited by 9 papers in the map (data through 7 Oct 2026). Builds on 13 of its 22 references that are in the map. First posted on arXiv 23 Sep 2024.

Builds on

13 of 22 references.

References · 13

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
  3. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,694
  4. 2020An Investigation of Why Overparameterization Exacerbates Spurious Correlationscited by 310
  5. 2023OBJECT 3DIT: Language-guided 3D-aware Image Editingcited by 62
  6. 2024TexSliders: Diffusion-Based Texture Editing in CLIP Spacecited by 23
  7. 2018Assessing Shape Bias Property of Convolutional Neural Networkscited by 18
  8. 2024Who's in and who's out? A case study of multimodal CLIP-filtering in DataCompcited by 15
  9. 2023The CLIP Model is Secretly an Image-to-Prompt Convertercited by 14
  10. 2024Isotropic3D: Image-to-3D Generation Based on a Single CLIP Embeddingcited by 13
  11. 2024Bag of Tricks to Boost Adversarial Transferabilitycited by 12
  12. 2024Reducing Texture Bias of Deep Neural Networks via Edge Enhancing Diffusioncited by 11
1 more
  1. 2024Discover and Mitigate Multiple Biased Subgroups in Image Classifierscited by 11

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 1; 2025: 5; 2026: 3.

9 papers.

  1. 2025Your Vision-Language Model Can't Even Count to 20: Exposing the Failures of VLMs in Compositional Countingcited by 17
  2. 2025On the Limitations of Vision-Language Models in Understanding Image Transformscited by 12
  3. 2025Targeted Forgetting of Image Subgroups in CLIP Modelscited by 6
  4. 2025Shape Bias and Robustness Evaluation via Cue Decomposition for Image Classification and Segmentationcited by 3
  5. 2025LLMs Can Compensate for Deficiencies in Visual Representationscited by 3
  6. 2024Will the Inclusion of Generated Data Amplify Bias Across Generations in Future Image Classification Models?cited by 1
  7. 2026Can Vision-Language Models See Squares? Text-Recognition Mediates Spatial Reasoning Across Three Model Familiescited by 1
  8. 2026Why Instruction-Based Unlearning Fails in Diffusion Models?cited by 0
  9. 2026Is CLIP Cross-Eyed? Revealing and Mitigating Center Bias in the CLIP Familycited by 0