IndiaCultureVL-Bench

cultural recognition and prompt-induced collapse in open vision-language models

A benchmark probing how open vision-language models recognize Indian cultural artifacts and contexts, and how sensitive their outputs are to prompt phrasing — including cases of prompt-induced collapse where small wording changes cause large drops in recognition accuracy.

(Sarkar, n.d.)

References

  1. IndiaCultureVL-Bench: A Prompt-Controlled Benchmark for Cultural Recognition in Open Vision-Language Models
    Suparnojit Sarkar
    Non archival publications at EMNLP 2026 ORACLE and ICML 2026 GlobalSouthML Workshops