Research map / Architectures and efficiency / Quantization
Soft then Hard: Rethinking the Quantization in Neural Image Compression
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 |
|---|---|---|---|
| soft-then-hard (STH) quantization strategy (ablation) | Cheng et al., 2020 base model · Kodak | RD performance (PSNR vs bpp) | Helped |
| scaled uniform noise (SUN) added to STH (ablation) | Cheng et al., 2020 base model · Kodak | RD performance (PSNR vs bpp): marginal | Helped |
| soft-then-hard plus scaled uniform noise combined | Cheng et al., 2020 base model · Kodak | BD-rate: 8.9% BD-rate savings | Helped |
| replacing additive uniform noise with straight-through estimator (STE) (ablation) | Cheng et al., 2020 base model · Kodak | RD performance (PSNR vs bpp) | Hurt |
| tuning AUN-pretrained model with stochastic Gumbel annealing (SGA) (ablation) | Cheng et al., 2020 base model · Kodak | RD performance (PSNR vs bpp) | Mixed |
| soft-then-hard (STH) quantization strategy (ablation) | Minnen et al., 2018 base model · Kodak | RD performance (PSNR vs bpp) | Helped |
| tuning AUN-pretrained model with SGA (ablation) | Minnen et al., 2018 base model · Kodak | convergence | Hurt |
| STE-based quantization (ablation) | Minnen et al., 2018 base model · Kodak | RD performance (PSNR vs bpp) | Hurt |
Closest papers by meaning
- Universally Quantized Neural Compression (NeurIPS 2020 · similarity 0.92)
- Robustly overfitting latents for flexible neural image compression (NeurIPS 2024 · similarity 0.91)
- Approaching Rate-Distortion Limits in Neural Compression with Lattice Transform Coding (ICLR 2025 · similarity 0.89)
- Bridging the Gap between Gaussian Diffusion Models and Universal Quantization for Image Compression (CVPR 2025 · similarity 0.88)
- Learning Optimal Lattice Vector Quantizers for End-to-end Neural Image Compression (NeurIPS 2024 · similarity 0.87)
- On Quantizing Neural Representation for Variable-Rate Video Coding (ICLR 2025 · similarity 0.87)
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.