Research map / Architectures and efficiency / Quantization
Non-uniform Step Size Quantization for Accurate Post-Training Quantization
Sangyun Oh, Hyeonuk Sim, Jounghyun Kim, Jongeun Lee
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 |
|---|---|---|---|
| subset quantization instead of BRECQ weight quantization (4-bit) | ResNet-18 · ImageNet W4/A32 | top-1 degradation (%): 0.38 -> 0.30 | Improved |
| subset quantization instead of BRECQ (3-bit weights) | ResNet-50 · ImageNet W3/A32 | top-1 degradation (%): 1.39 -> 0.99 | Improved |
| subset quantization instead of BRECQ (2-bit weights) | ResNet-18 · ImageNet W2/A32 | top-1 degradation (%): 4.78 -> 4.14 | Improved |
| subset quantization instead of BRECQ (2-bit weights) | ResNet-50 · ImageNet W2/A32 | top-1 degradation (%): 4.60 -> 3.86 | Improved |
| subset quantization instead of BRECQ with 4/4 quantization | ResNet-18 · ImageNet W4/A4 | top-1 degradation (%): 1.48 -> 1.21 | Improved |
| subset quantization instead of BRECQ with 4/4 quantization | ResNet-50 · ImageNet W4/A4 | top-1 degradation (%): 1.95 -> 1.48 | Hurt |
| subset quantization instead of BRECQ with 3/3 quantization | ResNet-18 · ImageNet W3/A3 | top-1 degradation (%): 5.05 -> 4.62 | Hurt |
| subset quantization instead of BRECQ with 3/3 quantization | ResNet-50 · ImageNet W3/A3 | top-1 degradation (%): 8.04 -> 6.80 | Hurt |
| subset quantization instead of BRECQ with 2/4 quantization | ResNet-18 · ImageNet W2/A4 | top-1 degradation (%): 6.28 -> 5.06 | Improved |
| subset quantization instead of PWLQ at 4-bit weights | DeepLabV3+ MobileNetV2 · segmentation W4/A8 | mIoU degradation (%): 3.15 -> 1.85 | Improved |
| subset quantization instead of uniform at 4-bit weights | DeepLabV3+ · segmentation W4/A8 | mIoU degradation (%): 20.76 -> 1.85 | Improved |
| subset quantization instead of uniform at 4-bit weights | SSD-Lite · detection W4/A8 | mAP degradation (%): 3.91 -> 0.38 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BRECQ — BRECQ: Pushing the Limit of Post-Training Quantization by Block Reconstruction (ICLR 2021) · better on 10 of 13 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Pushing the Limit of Post-Training Quantization (TPAMI 2025 · similarity 0.91)
- PTMQ: Post-training Multi-Bit Quantization of Neural Networks (AAAI 2024 · similarity 0.90)
- Exploring Post-training Quantization in LLMs from Comprehensive Study to Low Rank Compensation (AAAI 2024 · similarity 0.88)
- AIQViT: Architecture-Informed Post-Training Quantization for Vision Transformers (AAAI 2025 · similarity 0.88)
- Reg-PTQ: Regression-specialized Post-training Quantization for Fully Quantized Object Detector (CVPR 2024 · similarity 0.88)
- VLM-PTQ: Efficient Post-Training Quantization for Large Vision-Language Models (CVPR 2026 · similarity 0.88)
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.
