Research map / Label-efficient and robust learning / Knowledge distillation
Collaborative Distillation for Ultra-Resolution Universal Style Transfer
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the original encoder with Collaborative Distillation | VGG-19 in WCT · WCT user study | preference: 33.0% -> 42.7% | Helped |
| Replace the original encoder with Collaborative Distillation | VGG-19 in AdaIN · AdaIN user study | preference: 51.1% -> 42.4% | Mixed |
| Replace the original encoder with Collaborative Distillation | VGG-19 in Gatys style transfer · Gatys user study | preference: 46.5% -> 31.2% | Hurt |
| Use Collaborative Distillation instead of filter pruning | VGG-19 in WCT · WCT style distance, Conv1–Conv5 | style distance: 43.8 -> 28.1; 64.3 -> 43.8; 416.4 -> 269.7; 597.8 -> 446.4; 482.1 -> 399.0 | Helped |
| Compress the encoder using Collaborative Distillation | VGG-19 in WCT · 3000×3000 input | parameters: 17.1 -> 1.1(15.5×) | Helped |
| Compress the encoder using Collaborative Distillation | VGG-19 in WCT · 3000×3000 input | GPU inference time: 31.2 -> 8.6(3.6×) | Helped |
| Compress the encoder using Collaborative Distillation | VGG-19 in WCT · Square input on a 12GB GPU | maximum resolution: 3000×3000 -> 6000×6000 | Helped |
| Use linear embedding loss alone (ablation) | Collaborative Distillation for WCT · WCT stylization | visual quality | Hurt |
| Use collaborative loss alone (ablation) | Collaborative Distillation for WCT · WCT stylization | visual quality | Mixed |
| Use both linear embedding and collaborative losses (ablation) | Collaborative Distillation for WCT · WCT stylization | visual quality | Helped |
| Apply knowledge distillation to the decoder (ablation) | WCT decoder · WCT stylization | visual quality | Hurt |
| Raise classification top-5 accuracy from 80.55% to 83.47% with pruning fine-tuning | FP-pruned VGG-19 in WCT · WCT stylization | stylization quality: 80.55% -> 83.47% | No clear effect |
Closest papers by meaning
- MicroAST: Towards Super-fast Ultra-Resolution Arbitrary Style Transfer (AAAI 2023 · similarity 0.89)
- Towards Ultra-Resolution Neural Style Transfer via Thumbnail Instance Normalization (AAAI 2022 · similarity 0.88)
- StyleBank: An Explicit Representation for Neural Image Style Transfer (CVPR 2017 · similarity 0.87)
- Universal Style Transfer via Feature Transforms (NeurIPS 2017 · similarity 0.87)
- StyTr2: Image Style Transfer With Transformers (CVPR 2022 · similarity 0.87)
- Attention Distillation: A Unified Approach to Visual Characteristics Transfer (CVPR 2025 · similarity 0.85)
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.