Locate-then-Sparsify: Attribution Guided Sparse Strategy for Visual Hallucination Mitigation

Documenti analoghi

• Kernelized Sparse Fine-Tuning with Bi-level Parameter Competition for Vision Models
  di: Shen, Shufan, et al.
  Pubblicazione: (2025)
• Expanding Sparse Tuning for Low Memory Usage
  di: Shen, Shufan, et al.
  Pubblicazione: (2024)
• VL-SAE: Interpreting and Enhancing Vision-Language Alignment with a Unified Concept Set
  di: Shen, Shufan, et al.
  Pubblicazione: (2025)
• Enhancing Pre-trained Representation Classifiability can Boost its Interpretability
  di: Shen, Shufan, et al.
  Pubblicazione: (2025)
• VGS-Decoding: Visual Grounding Score Guided Decoding for Hallucination Mitigation in Medical VLMs
  di: Kolli, Govinda, et al.
  Pubblicazione: (2026)