Research map / Label-efficient and robust learning / Self-supervised and contrastive
Compressive Visual Representations
Kuang-Huei Lee, Anurag Arnab, Sergio Guadarrama, John Canny, Ian Fischer
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 CEB compression to SimCLR | SimCLR · ImageNet, ResNet-50, 1000 epochs | Top-1 accuracy: 70.7 -> 71.6 | Improved |
| Add CEB compression to BYOL | BYOL · ImageNet, ResNet-50, 1000 epochs | Top-1 accuracy: 74.2 -> 75.6 | Improved |
| Add CEB compression to BYOL | BYOL · ImageNet-C | Top-1 accuracy: 42.7 -> 45.1 | Improved |
| Add CEB compression to SimCLR | SimCLR · ImageNet, 1% labels | Top-1 accuracy: 49.3 -> 51.1 | Improved |
| Add CEB compression to BYOL | BYOL · ImageNet, 1% labels | Top-1 accuracy: 55.5 -> 60.6 | Improved |
| Increase compression strength to β = 1 (ablation) | C-SimCLR · ImageNet, 300 epochs | Top-1 accuracy: 69.0 -> 70.2 | Improved |
| Use β = 0 (no explicit compression) with a stochastic representation (ablation) | C-SimCLR · ImageNet, 300 epochs | Top-1 accuracy: 69.0 -> 69.7 | Improved |
| Increase compression strength beyond β = 1 to β = 1.5 (ablation) | C-SimCLR · ImageNet, 300 epochs | Top-1 accuracy: 70.2 -> 69.7 | Hurt |
| Increase compression strength to β = 2 (ablation) | C-SimCLR · ImageNet | training stability | Hurt |
| Increase BYOL loss weight from 1.0 to 5.0 (ablation) | BYOL · ImageNet, 300 epochs | Top-1 accuracy: 72.5 -> 72.8 | Improved |
| Increase BYOL loss weight from 1.0 to 5.0 (ablation) | BYOL · ImageNet, 1000 epochs | Top-1 accuracy: 74.2 -> 74.2 | No clear effect |
| Add a 256-d linear layer and l2-normalization (ablation) | BYOL · ImageNet, 300 epochs | Top-1 accuracy: 72.8 -> 72.8 | No clear effect |
Closest papers by meaning
- Benchmarking and Enhancing VLM for Compressed Image Understanding (ICML 2026 · similarity 0.88)
- From Flat to Hierarchical: Extracting Sparse Representations with Matching Pursuit (NeurIPS 2025 · similarity 0.87)
- Beyond Matryoshka: Revisiting Sparse Coding for Adaptive Representation (ICML 2025 · similarity 0.87)
- Vision Model Pre-training on Interleaved Image-Text Data via Latent Compression Learning (NeurIPS 2024 · similarity 0.86)
- Good, Cheap, and Fast: Overfitted Image Compression with Wasserstein Distortion (CVPR 2025 · similarity 0.86)
- Prune Redundancy, Preserve Essence: Vision Token Compression in VLMs via Synergistic Importance-Diversity (ICLR 2026 · similarity 0.86)
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.
