AI Paper Atlas

Research map / Architectures and efficiency / Quantization

Memory-Efficient Generative Models via Product Quantization

Jie Shao, Hanxiao Zhang, Hao Yu, Jianxin Wu

ICCV 2025 · Architectures and efficiency › Quantization · approach: Post-training weight quantization · 2 citations (OpenAlex)

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 changedWhereMeasuredEffect
use product quantization instead of vector quantization at 2-bitDiT-XL/2 · ImageNet 256x256, 250 stepsFID: 12.34 -> 6.84Improved
use product quantization instead of vector quantization at 1-bitDiT-XL/2 · ImageNet 256x256, 250 stepsFID: 91.71 -> 14.03Improved
use product quantization instead of vector quantization at 4-bitDiT-XL/2 · ImageNet 256x256, 250 stepsFID: 6.26 -> 6.25No clear effect
extend VQ calibration from 1 to 10 epochs (ablation)VQ 2-bit DiT-XL/2 · ImageNet 256x256, 250 stepsFID: 12.33 -> 12.12No clear effect
apply product quantization with full method vs GPTQ at 2-bitDiT-XL/2 · ImageNet 256x256, 250 stepsFID: 280.11 -> 6.84Improved
use our compression instead of Q-Diffusion at 2-bitLDM-4 · ImageNet 256x256, 20 stepsFID: 43.96 -> 7.98Improved
use our compression instead of SVDQuant at 2-bitPixArt-alpha · GenEval overallGenEval Overall: 0.17 -> 0.46Improved
use our compression instead of SVDQuant at 2-bitFLUX.1-dev · GenEval overallGenEval Overall: 0.32 -> 0.57Improved
apply our PQ method on top of Compact3DCompact3D · Tanks and TemplesPSNR: 23.39 -> 23.57Improved
apply our PQ method on top of Quip#Quip# LLaMA-2 7B · Wino/PiQA/Hellaaccuracy: Wino 62.43 -> 64.87Mixed
full method (PQ with DDPM loss, assignment adjustment) in ablation (ablation)Ours 2-bit DiT-XL/2 · ImageNet 256x256FID: 7.21 -> 6.84Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy