Research map / Architectures and efficiency / Quantization
Fast Lossless Neural Compression with Integer-Only Discrete Flows
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 |
|---|---|---|---|
| Use integer-only 8-bit inference with pruning instead of full-precision IDF-Dense | IDF compressor · ImageNet32, batch size 16 | inference latency (ms/sample): 4.08 -> 0.54 | Helped |
| Use integer-only 8-bit inference with pruning instead of full-precision IDF-Dense | IDF compressor · ImageNet32 | coding bpd: 3.900 -> 3.979 | Hurt |
| Use integer-only 8-bit inference with pruning instead of full-precision IDF-Dense | IDF compressor · ImageNet64, batch size 16 | inference latency (ms/sample): 13.93 -> 1.34 | Helped |
| Use integer-only 8-bit inference with pruning instead of full-precision IDF-Dense | IDF compressor · ImageNet64 | coding bpd: 3.638 -> 3.695 | Hurt |
| Replace DenseNet blocks with residual blocks (ablation) | IDF · ImageNet32, batch size 16 | inference latency (ms/sample): 4.08 -> 3.59 | Helped |
| Replace DenseNet blocks with residual blocks (ablation) | IDF · ImageNet32 | bpd: 3.890 -> 3.916 | Hurt |
| Replace DenseNet blocks with residual blocks (ablation) | IDF · ImageNet64, batch size 16 | inference latency (ms/sample): 13.93 -> 9.30 | Helped |
| Replace DenseNet blocks with residual blocks (ablation) | IDF · ImageNet64 | bpd: 3.629 -> 3.630 | Hurt |
| Quantize IDF-Res to 8-bit integer arithmetic (ablation) | IDF-Res · ImageNet32, batch size 16 | inference latency (ms/sample): 3.59 -> 0.64 | Helped |
| Quantize IDF-Res to 8-bit integer arithmetic (ablation) | IDF-Res · ImageNet32 | analytic bpd: 3.916 -> 3.923 | Hurt |
| Quantize IDF-Res to 8-bit integer arithmetic (ablation) | IDF-Res · ImageNet64, batch size 16 | inference latency (ms/sample): 9.30 -> 1.61 | Helped |
| Quantize IDF-Res to 8-bit integer arithmetic (ablation) | IDF-Res · ImageNet64 | analytic bpd: 3.637 -> 3.663 | Hurt |
Closest papers by meaning
- IDF++: Analyzing and Improving Integer Discrete Flows for Lossless Compression (ICLR 2021 · similarity 0.90)
- Integer Discrete Flows and Lossless Compression (NeurIPS 2019 · similarity 0.89)
- iFlow: Numerically Invertible Flows for Efficient Lossless Compression via a Uniform Coder (NeurIPS 2021 · similarity 0.87)
- Approaching Rate-Distortion Limits in Neural Compression with Lattice Transform Coding (ICLR 2025 · similarity 0.86)
- Optimal Neural Compressors for the Rate-Distortion-Perception Tradeoff (NeurIPS 2025 · similarity 0.85)
- Towards Lossless Implicit Neural Representation via Bit Plane Decomposition (CVPR 2025 · similarity 0.83)
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.