Research map / Vision-language and multimodal / CLIP and zero-shot
TIACam: Text-Anchored Invariant Feature Learning with Auto-Augmentation for Camera-Robust Zero-Watermarking
Abdullah Tanvir, Agnibh Dasgupta, Xin Zhong
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 the TIACam invariant feature extractor on top of CLIP (ablation) | CLIP · Feature invariance (Visual Genome) | cosine similarity: 0.78 -> 0.92 | Improved |
| add the TIACam invariant feature extractor on top of CLIP (ablation) | CLIP · Feature invariance (Flickr) | cosine similarity: 0.84 -> 0.93 | Improved |
| add the TIACam invariant feature extractor on top of CLIP (ablation) | CLIP · Feature invariance (MSCOCO) | cosine similarity: 0.76 -> 0.89 | Improved |
| add the TIACam invariant feature extractor on top of CLIP (ablation) | CLIP · Feature invariance (ImageNet) | cosine similarity: 0.82 -> 0.93 | Improved |
| use learned auto-augmentor instead of fixed manual distortions (ablation) | TIACam · Six random distortions (30-bit) | bit accuracy (%): 91.2 -> 98.7 | Improved |
| use learned auto-augmentor instead of fixed manual distortions (ablation) | TIACam · Six random distortions (100-bit) | bit accuracy (%): 89.5 -> 96.4 | Improved |
| use learned auto-augmentor instead of fixed manual distortions (ablation) | TIACam · Six random distortions (200-bit) | bit accuracy (%): 89.3 -> 96.1 | Improved |
| use learned auto-augmentor instead of fixed manual distortions (ablation) | TIACam · Feature invariance (Screen camera) | cosine similarity: 0.82 -> 0.91 | Improved |
| use learned auto-augmentor instead of fixed manual distortions (ablation) | TIACam · Feature invariance (Print camera) | cosine similarity: 0.76 -> 0.87 | Improved |
| use learned auto-augmentor instead of fixed manual distortions (ablation) | TIACam · Feature invariance (Screenshot) | cosine similarity: 0.78 -> 0.85 | Improved |
| use learned auto-augmentor instead of fixed manual distortions (ablation) | TIACam · Print camera (100-bit message) | bit accuracy (%): 87.4% -> 95.3% | Improved |
| use learned auto-augmentor instead of fixed manual distortions (ablation) | TIACam · Screen camera (30-bit message) | bit accuracy (%): 93.3% -> 99.1% | Improved |
Closest papers by meaning
- Rel-Zero: Harnessing Patch-Pair Invariance for Robust Zero-Watermarking Against AI Editing (CVPR 2026 · similarity 0.86)
- Disentangled Style Domain for Implicit $z$-Watermark Towards Copyright Protection (NeurIPS 2024 · similarity 0.84)
- Robust Watermarking Using Generative Priors Against Image Editing: From Benchmarking to Advances (ICLR 2025 · similarity 0.83)
- Lightweight-Mark: Rethinking Deep Learning-Based Watermarking (ICML 2025 · similarity 0.83)
- END^2: Robust Dual-Decoder Watermarking Framework Against Non-Differentiable Distortions (AAAI 2025 · similarity 0.82)
- RoPaSS: Robust Watermarking for Partial Screen-Shooting Scenarios (AAAI 2025 · similarity 0.82)
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.
