AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

Chain-of-Models Pre-Training: Rethinking Training Acceleration of Vision Foundation Models

Jiawei Fan, Shigeng Wang, Chao Li, Xiaolong Liu, Anbang Yao

CVPR 2026 · Architectures and efficiency › Pruning and efficient tuning · approach: Low-rank compression and merging · 1 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
add inverse weight initialization (IWI) to CoM-PT at 24 epochs (ablation)CoM-PT ViT-S · ImageNet-1Ktop-1: 26.48 -> 28.03Improved
add inverse feature distillation (IFD) to CoM-PT at 24 epochs (ablation)CoM-PT ViT-S · ImageNet-1Ktop-1: 26.48 -> 29.29Improved
combine IWI and IFD (ablation)CoM-PT ViT-S · ImageNet-1Ktop-1: 28.03 -> 30.24Improved
combine IWI and IFD (ablation)CoM-PT ViT-S · COCO retrievalR@1: 30.44 -> 34.15Improved
use the model chain (initialize from predecessor with distillation) instead of IWI from ViT-T only (ablation)ViT-B · ImageNet-1Ktop-1 / epochs: 92 -> 18 epochs; 31.71 -> 31.83Improved
use the model chain instead of IWI from ViT-T only (ablation)ViT-S · ImageNet-1Ktop-1 / epochs: 32 -> 24 epochs; 30.11 -> 30.24Improved
lengthen baseline pretraining from 32 to 128 epochsLaCLIP baseline · ImageNet-1K (CC3M)top-1: 28.80 -> 31.80Improved
replace baseline pretraining with CoM-PTViT-L/16 CC3M · COCOR@1: 36.53 -> 39.18Improved
replace baseline pretraining with CoM-PTSwin-L CC3M · ImageNet-1Ktop-1: 36.77 -> 37.06Improved
replace baseline backbone with CoM-PT backbone in SANViT-B/16 · Open-vocab segmentation averagemIoU: 23.53 -> 29.17Improved
replace baseline backbone with CoM-PT backbone in SANViT-L/16 · Open-vocab segmentation averagemIoU: 24.57 -> 30.71Improved
replace baseline backbone with CoM-PT backbone in LLaVA-1.5-7BViT-B/16 · ScienceQAaccuracy: 62.57 -> 63.41No clear effect

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