MAGIC: Margin-gated Contrastive Debiasing with Evidence Restoration for VLM Test-time Robustness Scaling
arXiv:2610.32013v1 Announce Type: new Abstract: This paper studies the problem of test-time robustness scaling for large vision-language models (LVLMs), which aims to mitigate language bias during inference. Existing approaches typically adopt test-time scaling techniques such as contrastive decodi…
Read the full story at arXiv cs.CV ↗
Timeline · 1 report
- 2026-10-26 04:00 · arXiv cs.CV
MAGIC: Margin-gated Contrastive Debiasing with Evidence Restoration for VLM Test-time Robustness Scaling