Research map / Vision-language and multimodal / CLIP and zero-shot
Guiding Diffusion-based Reconstruction with Contrastive Signals for Balanced Visual Representation
Boyu Han, Qianqian Xu, Shilong Bao, Zhiyong Yang, Ruochen Cui, Xilin Zhao, Qingming Huang
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 |
|---|---|---|---|
| replace naive Lcon+Lrec sum with unified DCR loss (ablation) | CLIP ViT-L@224 · MMVP-VLM | accuracy: 22.96 -> 33.30 | Improved |
| replace naive sum with DCR loss (ablation) | CLIP ViT-L@224 · Zero-shot clustering | NMI/ACC/ARI: 0.72/0.65/0.50 -> 0.76/0.67/0.54 | Improved |
| use two-stage (projector alignment then encoder) instead of end-to-end (ablation) | DCR · MMVP-VLM | accuracy: 25.93 -> 33.30 | Improved |
| use two-stage instead of end-to-end (ablation) | DCR · Zero-shot clustering | NMI/ACC/ARI: 0.73/0.65/0.51 -> 0.76/0.67/0.54 | Improved |
| use SD-2.1 backbone vs other SD versions (incl. SD-XL) (ablation) | DCR · MMVP-VLM and clustering | accuracy, NMI | Mixed |
| apply DCR to CLIP | OpenAI CLIP ViT-L@224 · MMVP-VLM | accuracy: 19.2 -> 33.3 | Improved |
| apply DIVA (reconstruction only) to CLIP | OpenAI CLIP ViT-L@224 · Zero-shot clustering | ACC: 0.61 -> 0.60 | Hurt |
| apply GenHancer to CLIP | OpenAI CLIP ViT-L@224 · Zero-shot clustering | NMI: 0.71 -> 0.65 | Hurt |
| swap in DCR-enhanced CLIP in LLaVA-1.5 | LLaVA-1.5 Vicuna-7B · MMVP-MLLM | accuracy: 24.7 -> 31.3 | Improved |
| swap in DCR-enhanced CLIP in LLaVA-1.5 | LLaVA-1.5 Vicuna-7B · POPE adv / SciQA-IMG | accuracy: 85.8 vs 84.2 (POPE adv); 67.5 vs 66.8 SciQA, below un2CLIP 68.4 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GenHancer — GenHancer: Imperfect Generative Models are Secretly Strong Vision-Centric Enhancers (ICCV 2025) · better on 9 of 9 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
- Diffusion Feedback Helps CLIP See Better (ICLR 2025 · similarity 0.89)
- Diff-2-in-1: Bridging Generation and Dense Perception with Diffusion Models (ICLR 2025 · similarity 0.89)
- ORCA: Hunting Compositional Failures in Text-to-Image Diffusion (NeurIPS 2026 · similarity 0.88)
- Unleashing Diffusion Transformers for Visual Correspondence by Modulating Massive Activations (NeurIPS 2025 · similarity 0.88)
- Rethinking Contrastive Loss in CLIP Post-training: A Complementary Framework with Frozen Text Encoder (NeurIPS 2026 · similarity 0.87)
- Why DiT Models Underperform as Representation Learners without Long Skip Connections (NeurIPS 2026 · 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.
