Research map / Vision-language and multimodal / CLIP and zero-shot
CLIP Is Shortsighted: Paying Attention Beyond the First Sentence
Marc-Antoine Lavoie, Anas Mahmoud, Aldo Zaimi, Arsene Fansi Tchango, Steven L. Waslander
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 |
|---|---|---|---|
| remove summary sentence from the short training caption (ablation) | Long-CLIP (3 epochs) · Urban1k T2I | top-1 recall: 82.7 -> 92.6 | Improved |
| remove summary sentence from the short training caption (ablation) | Long-CLIP (3 epochs) · DOCCI T2I | top-1 recall: 75.2 -> 80.9 | Improved |
| add sentence subsampling (ablation) | DeBias-CLIP w/o padding · Urban1k T2I | top-1 recall: 92.6 -> 92.5 | No clear effect |
| add token padding (ablation) | DeBias-CLIP w/ subsampling · COCO T2I | top-1 recall: 41.9 -> 43.0 | Improved |
| add token padding (ablation) | DeBias-CLIP w/ subsampling · DOCCI T2I | top-1 recall: 80.8 -> 79.7 | Hurt |
| DeBias-CLIP vs Long-CLIP | ViT-B/16 · DOCCI T2I | top-1 recall: 71.4 -> 80.0 | Improved |
| DeBias-CLIP vs SmartCLIP | ViT-B/16 · Urban1k T2I | top-1 recall: 87.4 -> 93.0 | Improved |
| DeBias-CLIP vs SmartCLIP | ViT-L/14 · COCO T2I | top-1 recall: 48.5 -> 48.1 | Hurt |
| train with DeBias-CLIP on original CLIP vs Long-CLIP under moved summary sentence | OpenAI CLIP · DOCCI T2I | T2I drop (Keep vs Move): -9.7% -> -3.5% | Improved |
| DeBias-CLIP on SigLIP/SigLIP2 under moved summary | SigLIP, SigLIP2 · DOCCI T2I | T2I drop (Keep vs Move): -6.1% and -6.5% | Mixed |
| increase short-caption loss weight above 0.25 (ablation) | DeBias-CLIP · COCO and DOCCI | T2I recall | Hurt |
| prepend padding sentences to captions (probe) (ablation) | CLIP, OpenCLIP, SigLIP, SigLIP2 · DOCCI T2I | top-1 retrieval | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- SmartCLIP — SmartCLIP: Modular Vision-language Alignment with Identification Guarantees (CVPR 2025) · better on 9 of 10 shared results
- Long-CLIP — Long-CLIP: Unlocking the Long-Text Capability of CLIP (ECCV 2024) · better on 6 of 6 shared results
- FineLIP — FineLIP: Extending CLIP's Reach via Fine-Grained Alignment with Longer Text Inputs (CVPR 2025) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CLIP Under the Microscope: A Fine-Grained Analysis of Multi-Object Representation (CVPR 2025 · similarity 0.92)
- Diffusion Feedback Helps CLIP See Better (ICLR 2025 · similarity 0.91)
- Modeling Caption Diversity in Contrastive Vision-Language Pretraining (ICML 2024 · similarity 0.90)
- Mining Fine-Grained Image-Text Alignment for Zero-Shot Captioning via Text-Only Training (AAAI 2024 · similarity 0.90)
- FineLIP: Extending CLIP's Reach via Fine-Grained Alignment with Longer Text Inputs (CVPR 2025 · similarity 0.90)
- un$^2$CLIP: Improving CLIP's Visual Detail Capturing Ability via Inverting unCLIP (NeurIPS 2025 · similarity 0.90)
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.
