Pith. sign in

REVIEW 2 cited by

When are Lemons Purple? The Concept Association Bias of Vision-Language Models

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2212.12043 v2 pith:EKOLYPL3 submitted 2022-12-22 cs.CV cs.CLcs.LG

When are Lemons Purple? The Concept Association Bias of Vision-Language Models

classification cs.CV cs.CLcs.LG
keywords modelsconceptperformancevision-languageassociationwhenzero-shotautoregressive
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Large-scale vision-language models such as CLIP have shown impressive performance on zero-shot image classification and image-to-text retrieval. However, such performance does not realize in tasks that require a finer-grained correspondence between vision and language, such as Visual Question Answering (VQA). As a potential cause of the difficulty of applying these models to VQA and similar tasks, we report an interesting phenomenon of vision-language models, which we call the Concept Association Bias (CAB). We find that models with CAB tend to treat input as a bag of concepts and attempt to fill in the other missing concept crossmodally, leading to an unexpected zero-shot prediction. We demonstrate CAB by showing that CLIP's zero-shot classification performance greatly suffers when there is a strong concept association between an object (e.g. eggplant) and an attribute (e.g. color purple). We also show that the strength of CAB predicts the performance on VQA. We observe that CAB is prevalent in vision-language models trained with contrastive losses, even when autoregressive losses are jointly employed. However, a model that solely relies on autoregressive loss seems to exhibit minimal or no signs of CAB.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Long Story Short: Disentangling Compositionality and Long-Caption Understanding in Contrastive VLMs

    cs.CV 2025-09 unverdicted novelty 5.0

    Empirical study shows bidirectional but sensitive relationship between compositionality and long-caption understanding in VLMs, promoted by high-quality grounded data and affected by architectural choices like frozen ...

  2. SRL-CLIP: Efficient CLIP Video Adaptation via Structured Semantic Role Labels

    cs.CV 2024-01 unverdicted novelty 5.0

    SRL-CLIP uses rule-based captions derived from semantic role labels to adapt CLIP via contrastive fine-tuning on 23k pairs, matching or exceeding larger models trained on far more data across video tasks.