Vision-language models are weak and prompt-sensitive at evoked emotion recognition, and many fine-grained errors are best explained by noisy dataset labels.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Evaluating Vision-Language Models for Emotion Recognition
Vision-language models are weak and prompt-sensitive at evoked emotion recognition, and many fine-grained errors are best explained by noisy dataset labels.