Research map / Architectures and efficiency / Quantization
PQ-SAM: Post-training Quantization for Segment Anything Model
Xiaoyu Liu, Xin Ding, Lei Yu, Yuanyuan Xi, Wei Li, Zhijun Tu, jie hu, Hanting Chen, Baoqun YIN, Zhiwei Xiong
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 |
|---|---|---|---|
| learn quantization step sizes and shifting and scaling sizes jointly (ablation) | PQ-SAM · SAM ViT-B point prompt BBBC (W4A4) | IoU: 0.6342 -> 0.7080 | Improved |
| make step sizes learnable instead of fixed (ablation) | PQ-SAM · SAM ViT-B point prompt BBBC (W4A4) | IoU: 0.0556 -> 0.6342 | Improved |
| group channels with outlier hierarchical clustering rather than per-channel parameters (ablation) | PQ-SAM ADT · SAM ViT-B point prompt BBBC (W4A4) | IoU: 0.6842 -> 0.7080 | Improved |
| group channels with OHC rather than evenly sized groups (ablation) | PQ-SAM ADT · SAM ViT-B point prompt BBBC (W4A4) | IoU: 0.6817 -> 0.7080 | Improved |
| add tensor-wise outlier truncating (ablation) | PQ-SAM · SAM ViT-B point prompt BBBC (W4A4) | IoU: 0.6342 -> 0.6533 | Improved |
| use PQ-SAM instead of RepQ-ViT for ViT-L at W4A4 | SAM point prompt · SAM ViT-L BBBC038V1 (W4A4) | IoU: 0.4129 -> 0.6855 | Improved |
| use PQ-SAM instead of RepQ-ViT on iShape for ViT-L at W4A4 | SAM point prompt · SAM ViT-L iShape (W4A4) | IoU: 0.2791 -> 0.2520 | Hurt |
| use PQ-SAM instead of Percentile on iShape for ViT-H at W6A6 | SAM point prompt · SAM ViT-H iShape (W6A6) | IoU: 0.3679 -> 0.3380 | Hurt |
| increase calibration data from 100 to 500 images (ablation) | PQ-SAM · SAM ViT-B point prompt BBBC (W4A4) | IoU: 0.7080 -> 0.7120 | No clear effect |
| change truncation threshold lambda away from 10 (ablation) | PQ-SAM · SAM ViT-B point prompt BBBC (W4A4) | IoU: 0.7080 -> 0.6901 (lambda 6) / 0.6858 (lambda 14) | Hurt |
Closest papers by meaning
- PTQ4SAM: Post-Training Quantization for Segment Anything (CVPR 2024 · similarity 0.94)
- TinySAM: Pushing the Envelope for Efficient Segment Anything Model (AAAI 2025 · similarity 0.90)
- AIQViT: Architecture-Informed Post-Training Quantization for Vision Transformers (AAAI 2025 · similarity 0.88)
- SAM 2: Segment Anything in Images and Videos (ICLR 2025 · similarity 0.87)
- VLM-PTQ: Efficient Post-Training Quantization for Large Vision-Language Models (CVPR 2026 · similarity 0.87)
- BA-SAM: Scalable Bias-Mode Attention Mask for Segment Anything Model (CVPR 2024 · 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.
