Research map / Vision-language and multimodal / CLIP and zero-shot
Immuno-VLM: Immunizing Large Vision-Language Models via Generative Semantic Antibodies for Open-World Trustworthiness
Xiang Fang, Wanlong Fang, Wei Ji
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Immuno-VLM semantic antibody training to frozen CLIP | CLIP ViT-B/16 · ImageNet-O (Near-OOD) | AUROC: 66.8 -> 83.4 | Improved |
| add Immuno-VLM to MCM-style zero-shot baseline | CLIP ViT-B/16 · ImageNet-O (Near-OOD) | FPR95: 61.3 -> 28.5 | Improved |
| add Immuno-VLM adapter to frozen CLIP | CLIP ViT-B/16 · ImageNet-1K ID | accuracy: 78.2 -> 78.5 | No clear effect |
| train CSI-style objective instead of keeping zero-shot | CLIP ViT-B/16 · ImageNet-1K ID | accuracy: 78.2 -> 76.5 | Hurt |
| add Gaussian-noise adapter negatives (ablation) | Frozen CLIP · ImageNet-O (Near-OOD) | AUROC: 66.8 -> 68.2 (+1.4) | Improved |
| use random dictionary words as negatives (ablation) | Frozen CLIP · ImageNet-O (Near-OOD) | AUROC: 66.8 -> 72.5 (+5.7) | Improved |
| use LLM-generated antibodies without filter (ablation) | Frozen CLIP · ImageNet-O (Near-OOD) | AUROC: 66.8 -> 79.8 (+13.0) | Improved |
| add semantic filter and EVT thresholding to LLM antibodies (ablation) | Immuno-VLM · ImageNet-O (Near-OOD) | AUROC: 79.8 -> 83.4 | Improved |
| add semantic filter and EVT thresholding to LLM antibodies (ablation) | Immuno-VLM · ImageNet-1K ID | accuracy: 77.6 -> 78.5 | Improved |
| set margin m to 0.0 instead of 0.2 (ablation) | Immuno-VLM · OOD detection | AUROC: 76.2 | Hurt |
| raise margin m to 0.5 or above (ablation) | Immuno-VLM · ImageNet-1K ID | accuracy: -4.2% | Hurt |
| remove the upper similarity bound of the antibody filter (ablation) | Immuno-VLM · ImageNet-1K ID | accuracy: -12% | Hurt |
Closest papers by meaning
- VL-RewardBench: A Challenging Benchmark for Vision-Language Generative Reward Models (CVPR 2025 · similarity 0.85)
- Seeing Clearly, Reasoning Confidently: Plug-and-Play Remedies for Vision Language Model Blindness (CVPR 2026 · similarity 0.84)
- Hyperbolic Learning with Synthetic Captions for Open-World Detection (CVPR 2024 · similarity 0.84)
- The Neglected Tails in Vision-Language Models (CVPR 2024 · similarity 0.84)
- Synthesize Diagnose and Optimize: Towards Fine-Grained Vision-Language Understanding (CVPR 2024 · similarity 0.84)
- UniBench: Visual Reasoning Requires Rethinking Vision-Language Beyond Scaling (NeurIPS 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
