Research map / Architectures and efficiency / RNNs and normalization
Towards Image Understanding from Deep Compression Without Decoding
Read the paperCheck 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 |
|---|---|---|---|
| Classify compressed representations rather than decoded RGB images with matched-compute networks | cResNet-51 vs ResNet-50 · ImageNet ILSVRC2012 val, 0.635 bpp | Top-5 accuracy: 88.34 -> 87.85 | Hurt |
| Classify compressed representations rather than decoded RGB images with matched-compute networks | cResNet-51 vs ResNet-50 · ImageNet ILSVRC2012 val, 0.330 bpp | Top-5 accuracy: 86.25 -> 85.87 | Hurt |
| Classify compressed representations rather than decoded RGB images with matched-compute networks | cResNet-51 vs ResNet-50 · ImageNet ILSVRC2012 val, 0.0983 bpp | Top-5 accuracy: 78.52 -> 78.20 | Hurt |
| Segment compressed representations rather than decoded RGB images | cResNet-51-d vs ResNet-50-d · PASCAL VOC-2012 val, 0.635 bpp | mIoU: 62.97 -> 62.86 | No clear effect |
| Segment compressed representations rather than decoded RGB images | cResNet-51-d vs ResNet-50-d · PASCAL VOC-2012 val, 0.330 bpp | mIoU: 60.75 -> 61.12 | Helped |
| Segment compressed representations rather than decoded RGB images | cResNet-51-d vs ResNet-50-d · PASCAL VOC-2012 val, 0.0983 bpp | mIoU: 52.97 -> 54.62 | Helped |
| Use the shallower cResNet-39 instead of cResNet-51 (ablation) | cResNet classification · ImageNet ILSVRC2012 val, 0.635 bpp | Top-5 accuracy: 87.85 -> 87.47 | Hurt |
| Use the shallower cResNet-39 instead of cResNet-51 (ablation) | cResNet classification · ImageNet ILSVRC2012 val, 0.330 bpp | Top-5 accuracy: 85.87 -> 85.46 | Hurt |
| Use the shallower cResNet-39 instead of cResNet-51 (ablation) | cResNet classification · ImageNet ILSVRC2012 val, 0.0983 bpp | Top-5 accuracy: 78.20 -> 77.65 | Hurt |
| Use the shallower cResNet-39-d instead of cResNet-51-d (ablation) | cResNet segmentation · PASCAL VOC-2012 val, 0.635 bpp | mIoU: 62.86 -> 61.85 | Hurt |
| Use the shallower cResNet-39-d instead of cResNet-51-d (ablation) | cResNet segmentation · PASCAL VOC-2012 val, 0.330 bpp | mIoU: 61.12 -> 60.78 | Hurt |
| Use the shallower cResNet-39-d instead of cResNet-51-d (ablation) | cResNet segmentation · PASCAL VOC-2012 val, 0.0983 bpp | mIoU: 54.62 -> 53.51 | Hurt |
Closest papers by meaning
- Full Resolution Image Compression With Recurrent Neural Networks (CVPR 2017 · similarity 0.87)
- What's in the Image? Explorable Decoding of Compressed Images (CVPR 2021 · similarity 0.87)
- Adaptive Learned Image Compression with Graph Neural Networks (CVPR 2026 · similarity 0.87)
- Distribution-Aware Tensor Decomposition for Compression of Convolutional Neural Networks (NeurIPS 2025 · similarity 0.86)
- GLIC: General Format Learned Image Compression (AAAI 2025 · similarity 0.86)
- Benchmarking and Enhancing VLM for Compressed Image Understanding (ICML 2026 · 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.