ClipTTT: CLIP-Guided Test-Time Training Helps LVLMs See Better
This addresses hallucination issues in LVLMs for real-world applications with visual corruptions, representing an incremental improvement.
The paper tackles the problem of hallucination in large vision-language models when visual inputs are corrupted at test time, showing that ClipTTT, a CLIP-guided test-time training method, effectively mitigates hallucinations and improves descriptive faithfulness, as demonstrated in experiments with 15 common corruptions.
Large vision-language models (LVLMs) tend to hallucinate, especially when visual inputs are corrupted at test time. We show that such corruptions act as additional distribution shifts, significantly amplifying hallucination rates in real-world applications. To address this, we propose CLIP-guided Test-Time Training (ClipTTT), a method to adapt LVLMs under degraded conditions on the fly with a single test sample. Specifically, we leverage the image-text alignment strength of a pre-trained CLIP model as a stable guidance signal to identify reliable self-supervision targets, enabling rapid adaptation without altering the base LVLMs. Extensive experiments on standard hallucination benchmarks, with 15 common corruptions, demonstrate that ClipTTT effectively mitigates hallucinations and improves descriptive faithfulness under visual corruptions.