Research map / Vision-language and multimodal / CLIP and zero-shot
Omniview-Tuning: Boosting Viewpoint Invariance of Vision-Language Pre-training Models
Shouwei Ruan, Yinpeng Dong, Liu Hanqing, Yao Huang, Hang Su, Xingxing Wei
Read the paperCheck your idea against 137,217 papers
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 cross-viewpoint alignment loss L_VC on top of L_ITC and VIFormer (ablation) | OVT-OpenCLIP ViT-L/14 · Viewpoint-OOD avg | Avg Top-1 accuracy: 57.5 -> 62.3 | Improved |
| add VIFormer module (ablation) | OVT-OpenCLIP ViT-L/14 · Viewpoint-OOD avg | Avg Top-1 accuracy: 56.0 -> 57.5 | Improved |
| fine-tune with only image-text contrastive loss on multi-view data (ablation) | OpenCLIP ViT-L/14 · Clean avg | Avg Top-1 accuracy: 81.9 -> 79.4 | Hurt |
| add L_VC to retain clean accuracy (ablation) | OVT-OpenCLIP ViT-L/14 · Clean avg | Avg Top-1 accuracy: 79.9 -> 81.8 | Improved |
| minimax outlier selection instead of random outlier viewpoint sampling (ablation) | OVT-OpenCLIP ViT-B/32 · Viewpoint-OOD avg | Avg Top-1 accuracy: 51.5 -> 52.6 | Improved |
| apply OVT fine-tuning to CLIP ViT-B/32, B/16, L/14 | OpenCLIP · Viewpoint-OOD avg | Avg Top-1 accuracy: +9.6%, +10.2%, +8.9% | Improved |
| apply OVT fine-tuning | OpenCLIP · 2D-OOD avg | Avg Top-1 accuracy: -2.6%, -1.4%, -0.2% | Hurt |
| use OVT-CLIP as visual encoder instead of OpenAI CLIP in LLaVA-7b | LLaVA-7b · ImageNet-V+ captioning | Description accuracy beta@Adp.: 32.1 -> 41.0 | Improved |
| use adversarially trained robust CLIP (TeCoA, FARE) as LLaVA encoder | LLaVA-7b · ImageNet-V+ captioning | Description accuracy beta@Adp.: 32.1 -> 22.6 (TeCoA), 23.1 (FARE) | Hurt |
| increase lambda (weight of viewpoint consistency loss) (ablation) | OVT-OpenCLIP ViT-B/32 · Clean and 2D-OOD avg | Avg Top-1 accuracy | Mixed |
| use OVT-CLIP in OpenFlamingo-3B | OpenFlamingo-3B · OOD-CV Pose beta@0.5 | VQA accuracy: 87.0 -> 82.5 | Mixed |
Closest papers by meaning
- OpenView: Empowering MLLMs with Out-of-view VQA (NeurIPS 2026 · similarity 0.88)
- UniBench: Visual Reasoning Requires Rethinking Vision-Language Beyond Scaling (NeurIPS 2024 · similarity 0.88)
- Adaptive Parameter Selection for Tuning Vision-Language Models (CVPR 2025 · similarity 0.87)
- Synthesize Diagnose and Optimize: Towards Fine-Grained Vision-Language Understanding (CVPR 2024 · similarity 0.87)
- Benchmarking Large Vision-Language Models on Fine-Grained Image Tasks: A Comprehensive Evaluation (ICLR 2026 · similarity 0.87)
- Machine Vision Therapy: Multimodal Large Language Models Can Enhance Visual Robustness via Denoising In-Context Learning (ICML 2024 · similarity 0.87)
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.
