Research map / Detection and segmentation / Open-set and few-shot detection
Rate-Distortion Theory in Coding for Machines and Its Applications
Alon Harell, Yalda Foroutan, Nilesh Ahuja, Parual Datta, Bhavya Kanzariya, V. Srinivasa Somayazulu, Omesh Tickoo, Anderson de Andrade, Ivan V. Bajić
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 |
|---|---|---|---|
| measure MSE at deeper layer (Features.32) instead of the input | VGG16 · CIFAR-10 | task-appropriateness: 0.038 -> 1.0 | Improved |
| measure MSE at deeper layer (Layer4) instead of the input | ResNet50 · CIFAR-100 | task-appropriateness: 0.036 -> 0.997 | Improved |
| distil at intermediate layer Y instead of input X under 1-bit MSE quantisation | toy classifier · toy example | classification error: 50% -> perfect classification | Improved |
| use a deeper distillation point for unsupervised model-splitting coding (ablation) | ResNet50 classification codec · ImageNet val | rate-accuracy | Improved |
| use supervised task loss instead of unsupervised distillation | model-splitting codec · ImageNet val | rate-accuracy | Improved |
| distil the output of layers after the compact ResNet50 backbone (M5 instead of M4) (ablation) | Deeplab-v3 codec · COCO 2017 val | rate-mIoU | Hurt |
| move the cut point deeper in constrained model-splitting codec (ablation) | model-splitting codec · ImageNet val | rate-accuracy | Improved |
| change the distillation point from C5 to C6 (ablation) | Faster R-CNN direct codec · COCO2017 val | BD-rate: 36.57% BD-rate saving | Improved |
| also move the cut point to C6 (ablation) | Faster R-CNN direct codec · COCO2017 val | BD-rate: 2% more savings in BD-rate | No clear effect |
| use deepest distillation point M6 with cut C6 (ablation) | Faster R-CNN direct codec · COCO2017 val | BD-rate: 82.34% BD-rate savings, 4.38% BD-mAP | Improved |
Closest papers by meaning
- Unified Coding for Both Human Perception and Generalized Machine Analytics with CLIP Supervision (AAAI 2025 · similarity 0.87)
- Towards image compression with perfect realism at ultra-low bitrates (ICLR 2024 · similarity 0.86)
- All-in-One Image Coding for Joint Human-Machine Vision with Multi-Path Aggregation (NeurIPS 2024 · similarity 0.86)
- Test-Time Fine-Tuning of Image Compression Models for Multi-Task Adaptability (CVPR 2025 · similarity 0.86)
- Good, Cheap, and Fast: Overfitted Image Compression with Wasserstein Distortion (CVPR 2025 · similarity 0.85)
- Benchmarking and Enhancing VLM for Compressed Image Understanding (ICML 2026 · 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.
