Research map / Label-efficient and robust learning / Knowledge distillation
Learn More for Food Recognition via Progressive Self-Distillation
Yaohui Zhu, Linhu Liu, Jiang Tian
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 |
|---|---|---|---|
| Add progressive self-distillation to Swin-B | Swin-B · ETHZ Food-101 | Top1 accuracy: 93.91 -> 94.56 | Improved |
| Add progressive self-distillation to Swin-B | Swin-B · Vireo Food-172 | Top1 accuracy: 92.75 -> 92.91 | Improved |
| Add progressive self-distillation to Swin-B | Swin-B · ISIA Food-500 | Top1 accuracy: 67.32 -> 70.10 | Improved |
| Add progressive self-distillation to DenseNet161 | DenseNet161 · ETHZ Food-101 | Top1 accuracy: 86.93 -> 87.40 | Improved |
| Add progressive self-distillation to DenseNet161 | DenseNet161 · Vireo Food-172 | Top1 accuracy: 88.25 -> 89.00 | Improved |
| Add progressive self-distillation to DenseNet161 | DenseNet161 · ISIA Food-500 | Top1 accuracy: 60.05 -> 60.94 | Improved |
| Use 1 self-distillation instead of 2 (ablation) | Swin-B+PSD · ETHZ Food-101 | Top1 accuracy: 94.56 -> 94.46 | Hurt |
| Use 1 self-distillation instead of 2 (ablation) | Swin-B+PSD · ISIA Food-500 | Top1 accuracy: 70.10 -> 69.47 | Hurt |
| Add a third self-distillation to the two-stage method (ablation) | Swin-B+PSD · ETHZ Food-101 | Top1 accuracy: 94.56 -> 94.44 | Hurt |
| Add a third self-distillation to the two-stage method (ablation) | Swin-B+PSD · ISIA Food-500 | Top1 accuracy: 70.10 -> 69.94 | Hurt |
| Train the self-distillations step by step rather than jointly with ramp-up weighting (ablation) | Swin-B+PSD · ETHZ Food-101 | Top1 accuracy: 94.56 -> 94.40 | Hurt |
| Share the teacher and student head networks (ablation) | Swin-B+PSD · ISIA Food-500 | Top1 accuracy: 70.10 -> 69.79 | Hurt |
Closest papers by meaning
- Large Scale Visual Food Recognition (TPAMI 2023 · similarity 0.87)
- SDPGO: Efficient Self-Distillation Training Meets Proximal Gradient Optimization (NeurIPS 2025 · similarity 0.86)
- Learning Cross-Modal Embeddings for Cooking Recipes and Food Images (CVPR 2017 · similarity 0.86)
- Learning Metrics From Teachers: Compact Networks for Image Embedding (CVPR 2019 · similarity 0.86)
- Multi-label Self Knowledge Distillation (AAAI 2025 · similarity 0.85)
- OVFoodSeg: Elevating Open-Vocabulary Food Image Segmentation via Image-Informed Textual Representation (CVPR 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.
