Pith. sign in

REVIEW 9 cited by

Long-CLIP: Unlocking the Long-Text Capability of CLIP

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2403.15378 v3 pith:E7PSPX7P submitted 2024-03-22 cs.CV

classification cs.CV
keywords cliplong-cliptexttext-imageretrievallengthsignificantcapabilities
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Contrastive Language-Image Pre-training (CLIP) has been the cornerstone for zero-shot classification, text-image retrieval, and text-image generation by aligning image and text modalities. Despite its widespread adoption, a significant limitation of CLIP lies in the inadequate length of text input. The length of the text token is restricted to 77, and an empirical study shows the actual effective length is even less than 20. This prevents CLIP from handling detailed descriptions, limiting its applications for image retrieval and text-to-image generation with extensive prerequisites. To this end, we propose Long-CLIP as a plug-and-play alternative to CLIP that supports long-text input, retains or even surpasses its zero-shot generalizability, and aligns the CLIP latent space, making it readily replace CLIP without any further adaptation in downstream frameworks. Nevertheless, achieving this goal is far from straightforward, as simplistic fine-tuning can result in a significant degradation of CLIP's performance. Moreover, substituting the text encoder with a language model supporting longer contexts necessitates pretraining with vast amounts of data, incurring significant expenses. Accordingly, Long-CLIP introduces an efficient fine-tuning solution on CLIP with two novel strategies designed to maintain the original capabilities, including (1) a knowledge-preserved stretching of positional embedding and (2) a primary component matching of CLIP features. With leveraging just one million extra long text-image pairs, Long-CLIP has shown the superiority to CLIP for about 20% in long caption text-image retrieval and 6% in traditional text-image retrieval tasks, e.g., COCO and Flickr30k. Furthermore, Long-CLIP offers enhanced capabilities for generating images from detailed text descriptions by replacing CLIP in a plug-and-play manner.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 9 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. AnyStyle: Single-Pass Multimodal Stylization for 3D Gaussian Splatting

    cs.CV 2026-02 conditional novelty 6.0 of 10

    A single-pass 3D Gaussian splatting pipeline that stylizes unposed scenes from either a text prompt or a reference image via a lightweight zero-initialized style-injection branch.

  2. PoemTale Diffusion: Minimising Information Loss in Poem to Image Generation with Multi-Stage Prompt Refinement

    cs.CV 2025-07 conditional novelty 6.0 of 10

    PoemTale Diffusion generates a coherent set of images from a poem by combining emotion-based segmentation, multi-stage LLM prompt refinement, and consistent self-attention, outperforming direct poem-to-image approache...

  3. FIX-CLIP: Dual-Branch Hierarchical Contrastive Learning via Synthetic Captions for Better Understanding of Long Text

    cs.CV 2025-07 conditional novelty 6.0 of 10

    A dual-branch CLIP training pipeline with regional prompts and hierarchical feature alignment reaches state-of-the-art on long- and short-text retrieval.

  4. GenEscape: Hierarchical Multi-Agent Generation of Escape Room Puzzles

    cs.CV 2025-06 conditional novelty 6.0 of 10

    A hierarchical multi-agent framework with GPT-4o generates escape room puzzle images that are judged more solvable and less shortcut-prone than vanilla text-to-image outputs.

  5. JuZhou 1.0 Technical Report: The First Edge-Native Text-to-Image Foundation Model Trained Entirely on China-Developed AI Accelerators

    cs.CV 2026-06 unverdicted novelty 5.0 of 10

    JuZhou 1.0 is a 0.387B-parameter T2I diffusion model with 4-step inference achieving 0.69 GenEval, trained on 9M Chinese pairs using Sugon K100 accelerators and deployable on Android/iOS devices.

  6. Domain-Enhanced Dual-Branch Model for Efficient and Interpretable Accident Anticipation

    cs.CV 2025-07 conditional novelty 5.0 of 10

    The paper claims 87.7% AP on DAD by aligning VGG-16 video features with Long-CLIP embeddings of California DMV accident reports and GPT-generated safe-driving reports.

  7. ANT: Adaptive Neural Temporal-Aware Text-to-Motion Model

    cs.CV 2025-06 conditional novelty 5.0 of 10

    ANT makes text embeddings change across denoising steps and schedules classifier-free guidance to decay, improving text-motion alignment in diffusion text-to-motion models.

  8. Multiverse Through Deepfakes: The MultiFakeVerse Dataset of Person-Centric Visual and Conceptual Manipulations

    cs.MM 2025-06 conditional novelty 5.0 of 10

    MultiFakeVerse provides 845,286 person-centric images edited through VLM-generated instructions; state-of-the-art deepfake detectors and human observers misclassify a large fraction of them.

  9. Distill CLIP (DCLIP): Enhancing Image-Text Retrieval via Cross-Modal Transformer Distillation

    cs.CV 2025-05 conditional novelty 5.0 of 10

    DCLIP fine-tunes a CLIP student's image encoder to match a YOLO-region, bidirectional cross-attention teacher, improving retrieval while retaining most zero-shot accuracy.

Pith tools