Research map / Image generation / Editing and style transfer
Learning the Latent "Look": Unsupervised Discovery of a Style-Coherent Embedding From Fashion Images
Wei-Lin Hsiao, Kristen Grauman
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 |
|---|---|---|---|
| use polylingual LDA instead of monolingual LDA (ablation) | style topic model · HipsterWars (GT attributes) | avg max AP: 0.48 -> 0.53 | Improved |
| use polylingual LDA instead of monolingual LDA (ablation) | style topic model · HipsterWars (GT attributes) | NMI: 0.30 -> 0.31 | No clear effect |
| use ground truth attributes instead of predicted attributes (ablation) | PolyLDA · DeepFashion style discovery | avg AP: 0.0647 -> 0.1762 | Improved |
| use ground truth attributes instead of predicted attributes (ablation) | PolyLDA · HipsterWars style discovery | avg AP: 0.50 -> 0.53 | Improved |
| use PolyLDA topics instead of raw attribute indicator vectors (predicted attributes) | Attributes · HipsterWars style discovery | avg AP: 0.28 -> 0.50 | Improved |
| use PolyLDA topics instead of StyleNet features | StyleNet · DeepFashion style discovery | NMI: 0.0011 -> 0.0116 | Improved |
| fine-tune ResNet-50 on 148 attributes (Attr-ResNet) instead of vanilla ImageNet ResNet-50 | ResNet-50 · HipsterWars style discovery | avg AP: 0.30 -> 0.35 | Improved |
| fine-tune CNN on fashion metadata labels (StyleNet) instead of vanilla ResNet | ResNet-50 · HipsterWars style discovery | avg AP: 0.30 -> 0.39 | Improved |
| choose clustering algorithm (K-means, GMM, AP-clustering) for baseline embeddings (ablation) | baseline CNN features · style discovery | accuracy | No clear effect |
| vary number of topics K from 15 to 30 in style mixing (ablation) | PolyLDA · style mixing | AP | No clear effect |
| use PolyLDA instead of StyleNet for mixing styles (GT attributes) | StyleNet · Style mixing, Bohemian x Hipster | AP: 0.113 -> 0.394 | Improved |
| use PolyLDA with predicted attributes instead of Attributes baseline for Hipster x Goth mixing | Attributes · Style mixing, Hipster x Goth | AP: 0.172 -> 0.424 | Improved |
Closest papers by meaning
- Easy Regional Contrastive Learning of Expressive Fashion Representations (NeurIPS 2024 · similarity 0.89)
- Learning Type-Aware Embeddings for Fashion Compatibility (ECCV 2018 · similarity 0.89)
- Learning to Appreciate the Aesthetic Effects of Clothing (AAAI 2016 · similarity 0.88)
- Towards Better Understanding the Clothing Fashion Styles: A Multimodal Deep Learning Approach (AAAI 2017 · similarity 0.88)
- Learning Attribute and Class-Specific Representation Duet for Fine-Grained Fashion Analysis (CVPR 2023 · similarity 0.88)
- Investigating Style Similarity in Diffusion Models (ECCV 2024 · similarity 0.88)
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.
