Research map / Vision-language and multimodal / Human and embodied modeling
Kaleido-BERT: Vision-Language Pre-Training on Fashion Domain
Mingchen Zhuge, Dehong Gao, Deng-Ping Fan, Linbo Jin, Ben Chen, Haoming Zhou, Minghui Qiu, Ling Shao
Check 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 |
|---|---|---|---|
| Use kaleido patch generation instead of fixed-scale patches (ablation) | Kaleido-BERT · Fashion-Gen | Sum R: 276.97 -> 303.80(+9.7%) | Improved |
| Add salient object detection to kaleido patch generation (ablation) | Kaleido-BERT · Fashion-Gen | Sum R: 276.97 -> 319.52(+16.2%) | Improved |
| Use alignment-guided masking instead of random masking (ablation) | Kaleido-BERT · Fashion-Gen | Sum R: 299.50 -> 319.52(+6.7%) | Improved |
| Add rotation recognition to AMLM and ITM (ablation) | Kaleido-BERT · Fashion-Gen | Sum R: 295.70 -> 297.18(+0.5%) | Improved |
| Add jigsaw solving after rotation recognition (ablation) | Kaleido-BERT · Fashion-Gen | Sum R: 297.18 -> 300.82(+1.7%) | Improved |
| Add camouflage prediction after rotation and jigsaw tasks (ablation) | Kaleido-BERT · Fashion-Gen | Sum R: 300.82 -> 312.65(+5.7%) | Improved |
| Add grey-to-color modeling after the first three kaleido subtasks (ablation) | Kaleido-BERT · Fashion-Gen | Sum R: 312.65 -> 316.10(+6.9%) | Improved |
| Add blank-to-color modeling after the first four kaleido subtasks (ablation) | Kaleido-BERT · Fashion-Gen | Sum R: 316.10 -> 319.02(+7.9%) | Improved |
| Add blank-to-color modeling alone to AMLM and ITM (ablation) | Kaleido-BERT · Fashion-Gen | Sum R: 295.70 -> 287.70(-2.7%) | Hurt |
| Replace blank-to-color modeling alone with rotation recognition (ablation) | Kaleido-BERT · Fashion-Gen | Sum R: 287.70 -> 297.18(+0.5%) | Improved |
| Use Kaleido-BERT rather than FashionBERT | Fashion-Gen ITR · Fashion-Gen ITR | Rank@1: 23.96% -> 27.99% | Improved |
| Use Kaleido-BERT rather than FashionBERT | Fashion-Gen TIR · Fashion-Gen TIR | Rank@1: 26.75% -> 33.88% | Improved |
Later papers that compared with this method
2 later accepted papers report KaleidoBERT as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- FashionViL: Fashion-Focused Vision-and-Language Representation Learning (ECCV 2022 · better on 5 of 5 shared results)
- Single-Stream Multi-level Alignment for Vision-Language Pretraining (ECCV 2022 · better on 1 of 1 shared result)
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
- Easy Regional Contrastive Learning of Expressive Fashion Representations (NeurIPS 2024 · similarity 0.89)
- FashionViL: Fashion-Focused Vision-and-Language Representation Learning (ECCV 2022 · similarity 0.88)
- VL-BERT: Pre-training of Generic Visual-Linguistic Representations (ICLR 2020 · similarity 0.87)
- Seeing the Abstract: Translating the Abstract Language for Vision Language Models (CVPR 2025 · similarity 0.87)
- SyncMask: Synchronized Attentional Masking for Fashion-centric Vision-Language Pretraining (CVPR 2024 · similarity 0.86)
- WikiDO: A New Benchmark Evaluating Cross-Modal Retrieval for Vision-Language Models (NeurIPS 2024 · similarity 0.84)
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.
