AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

Parameter-Efficient and Memory-Efficient Tuning for Vision Transformer: A Disentangled Approach

Taolin Zhang, Jiawang Bai, Zhihe Lu, Dongze Lian, genping wang, Xinchao Wang, Shu-Tao Xia

ECCV 2024 · Architectures and efficiency › Pruning and efficient tuning · approach: Efficient ViT adaptation · 7 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
replace VPT-Deep with SynQT disentangled query-only tuningViT-B/16 ImageNet-21K · VTAB-1Ktraining memory (GB): 8.1 -> 3.4Improved
replace VPT-Deep with SynQTViT-B/16 ImageNet-21K · VTAB-1Kaverage Top-1 accuracy: 72.0 -> 72.9Improved
replace VQT with SynQTViT-B/16 ImageNet-21K · VTAB-1Kaverage Top-1 accuracy: 68.3 -> 72.9Improved
apply memory constraint by dropping structures near input; SynQT versus AdaptFormerViT-B/16 ImageNet-21K · VTAB-1K (5GB constraint)average Top-1 accuracy: 65.8 -> 72.9Improved
apply SynQT to different pre-training backbonesMAE, DINO, supervised ImageNet-1K ViT-B/16 · VTAB-1Kaverage Top-1 accuracy: MAE 67.1 -> 68.9; DINO 69.9 -> 72.3; supervised 70.6 -> 71.4Improved
apply SynQT to Swin-B and ConvNeXt-BSwin-B, ConvNeXt-B · VTAB-1Kaverage Top-1 accuracy: Swin 71.6 -> 73.8; ConvNeXt 68.7 -> 69.5Improved
add trainable prompt to QSM input (ablation)SynQT · VTAB-1Kaverage Top-1 accuracy: 70.7 -> 72.9Improved
feed previous block output into QSM (ablation)SynQT · VTAB-1Kaverage Top-1 accuracy: 71.4 -> 72.9Improved
add DropFeat regularization in classification head (ablation)SynQT · VTAB-1Kaverage Top-1 accuracy: 72.4 -> 72.9Improved
use conditional feature aggregation instead of simple average (ablation)SynQT head · VTAB-1Kaverage Top-1 accuracy: 72.7 -> 72.9Improved
use shared instead of independent projection in head (ablation)SynQT head · VTAB-1Kaverage Top-1 accuracy: 72.6 -> 72.9Improved
increase QSM hidden size beyond 48 (ablation)SynQT · VTAB-1Kaverage Top-1 accuracy: 72.9 -> 72.7 (width 96)No 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