Research map / Architectures and efficiency / Quantization
A Little Bit More: Bitplane-Wise Bit-Depth Recovery
Abhijith Punnappurath, Michael S. Brown
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 |
|---|---|---|---|
| predict bitplanes with BCE instead of predicting the next-bit image (ablation) | Ours D4 · TESTIMAGES 1200 4-8 bit | PSNR: 35.8441 -> 39.6822 | Improved |
| use BCE loss instead of VGG perceptual loss (ablation) | Ours D4 · TESTIMAGES 1200 4-8 bit | PSNR: 35.2064 -> 39.6822 | Improved |
| use BCE loss instead of MSE loss (ablation) | Ours D4 · TESTIMAGES 1200 4-8 bit | PSNR: 39.6185 -> 39.6822 | Improved |
| use bitplane-wise training instead of single-shot residual network of same capacity (ablation) | Ours D4 · TESTIMAGES 1200 4-8 bit | PSNR: 37.4822 -> 39.6822 | Improved |
| use ResNet-style network instead of U-Net per bitplane (D4 size) (ablation) | bitplane training · TESTIMAGES 1200 4-8 bit | PSNR: 38.5855 -> 39.6822 | Improved |
| increase depth from D4 to D16 (ablation) | Ours · Kodak 4-8 bit | PSNR: 39.4171 -> 39.5185 | Improved |
| increase depth beyond D=16 (ablation) | Ours · TESTIMAGES 1200 4-8 bit | PSNR | No clear effect |
| restore more bitplanes accumulating from 5th to 8th (ablation) | Ours D16 · TESTIMAGES 1200 4-8 bit | PSNR: 38.5700 -> 40.3906 | Improved |
| train jointly end-to-end instead of independently (ablation) | Ours D4 · Sintel 4-8 bit | PSNR: 40.6143 -> 40.1737 | Mixed |
| train on mix of natural and animated images instead of animated only (ablation) | Ours D4 · MIT-Adobe FiveK 4-8 bit | PSNR: 38.6983 -> 39.4520 | Improved |
| train on mix instead of natural only (ablation) | Ours D4 · Sintel 4-8 bit | PSNR: 39.9516 -> 40.6143 | Improved |
| use bitplane-wise training with D16 instead of BitNet | Ours D16 · Sintel 4-16 bit | PSNR: 39.4893 -> 41.5070 | Improved |
Closest papers by meaning
- Predicting High-precision Depth on Low-Precision Devices Using 2D Hilbert Curves (ICML 2025 · similarity 0.86)
- AdaBM: On-the-Fly Adaptive Bit Mapping for Image Super-Resolution (CVPR 2024 · similarity 0.86)
- QuartDepth: Post-Training Quantization for Real-Time Depth Estimation on the Edge (CVPR 2025 · similarity 0.84)
- Untrained Neural Nets for Snapshot Compressive Imaging: Theory and Algorithms (NeurIPS 2024 · similarity 0.84)
- Towards Lossless Implicit Neural Representation via Bit Plane Decomposition (CVPR 2025 · similarity 0.84)
- 2DQuant: Low-bit Post-Training Quantization for Image Super-Resolution (NeurIPS 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.
