Introduces Synergistic Faithfulness metric based on Shapley Interaction Index to evaluate cross-modal synergy in VLM explainers, revealing over-reliance on visual salience in existing methods.
Repope: Impact of annotation errors on the pope benchmark
3 Pith papers cite this work. Polarity classification is still indexing.
3
Pith papers citing it
abstract
Since data annotation is costly, benchmark datasets often incorporate labels from established image datasets. In this work, we assess the impact of label errors in MSCOCO on the frequently used object hallucination benchmark POPE. We re-annotate the benchmark images and identify an imbalance in annotation errors across different subsets. Evaluating multiple models on the revised labels, which we denote as RePOPE, we observe notable shifts in model rankings, highlighting the impact of label quality. Code and data are available at https://github.com/YanNeu/RePOPE .
years
2026 3representative citing papers
TRACE’s fixed-k cross-entropy mask search and COPAIR’s coarse-pair warm-start raise search-based visual attribution faithfulness and enable high single-point RePOPE repair rates.
VLMs show language-prior reliance via multi-stage bottlenecks in visual retrieval and suppression, with many benchmark examples remaining answerable under severe visual obfuscation.
citing papers explorer
-
Measuring Cross-Modal Synergy: A Benchmark for VLM Explainability
Introduces Synergistic Faithfulness metric based on Shapley Interaction Index to evaluate cross-modal synergy in VLM explainers, revealing over-reliance on visual salience in existing methods.
-
A Good Initialization is All You Need for Faithful Visual Attribution
TRACE’s fixed-k cross-entropy mask search and COPAIR’s coarse-pair warm-start raise search-based visual attribution faithfulness and enable high single-point RePOPE repair rates.
-
Diagnosing Visual Ignorance in Vision-Language Models
VLMs show language-prior reliance via multi-stage bottlenecks in visual retrieval and suppression, with many benchmark examples remaining answerable under severe visual obfuscation.