AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

Self-Supervised Weight Templates for Scalable Vision Model Initialization

Yucheng Xie, Fu Feng, Ruixiao Shi, Jing Wang, Yong Rui, Xin Geng

ICML 2026 · Architectures and efficiency › Pruning and efficient tuning · approach: Low-rank compression and merging · 0 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 Tucker-based constraint instead of no constraint in pre-training (ablation)SWEET · ImageNet classification, L5H3Top-1 accuracy: 54.51 -> 56.70Improved
use Tucker constraint instead of Kronecker constraint (ablation)SWEET · ImageNet classification, L5H3Top-1 accuracy: 55.88 -> 56.70Improved
use Tucker constraint instead of linear factorization (ablation)SWEET · ImageNet classification, L5H3Top-1 accuracy: 55.66 -> 56.70Improved
add width-wise stochastic scaling during pre-training (ablation)SWEET · ImageNet classification, L3H4Top-1 accuracy: 53.92 -> 55.21Improved
add width-wise stochastic scaling during pre-training (ablation)SWEET · ImageNet classification, L3H5Top-1 accuracy: 58.65 -> 58.80Improved
add SwiGLU, RMSNorm and RoPE architectural enhancements (ablation)SWEET · ImageNet classification, L3H5Top-1 accuracy: 57.80 -> 58.80Improved
replace classification-trained templates (WAVE) with self-supervised Tucker templates (SWEET)weight-template initialization · ImageNet classification, average over 5 sizesTop-1 accuracy: 67.19 -> 68.79Improved
use SWEET instead of WAVE for detection initializationViT detectors · COCO detection, averageAPbox: 29.55 -> 31.59Improved
use SWEET instead of WAVE for segmentation initializationViT segmenters · ADE20K, averagemIoU: 30.32 -> 33.08Improved
use SWEET instead of DMAE for generation initializationViT generators · ImageNet generation, averageFID: 28.86 -> 26.67Improved
use SWEET instead of baselines on small fine-grained datasetsViT L6H6 · Oxford FlowerTop-1 accuracy: 83.18 -> 83.12No clear effect
use SWEET instead of WAVE for ConvNeXt-v2 initializationConvNeXt-v2 atto-L4 · ImageNet classificationTop-1 accuracy: 57.20 -> 57.31No clear effect

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.

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

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