Reproduce “Beyond Similarity Entropy: A Rank-Guided Data-Free Quantization Framework for Vision Transformers” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Runqing Jiang, Ye Zhang, Pengpeng Yu, Zhuo Song, Changjun Li, Longguang Wang, Yulan Guo

RankQ-ViT develops data-free quantization calibration for vision transformers using synthetic images generated from Gaussian noise. It promotes patch-feature diversity with a block-aware self-contrastive objective, then selects samples by feature rank and semantic consistency before block-wise reconstruction. On ImageNet DeiT-T at W4/A4, using 32 synthetic images and no real images, it reports 44.35% Top-1 accuracy versus 20.20% for PSAQ-ViT v2.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.