Papers › ELSA: Enhanced Local Self-Attention for Vision Transformer

ELSA: Enhanced Local Self-Attention for Vision Transformer

23 Dec 2021arXiv:2112.12786archive 2025-07-28

Jingkai Zhou, Pichao Wang, Fan Wang, Qiong Liu, Hao Li, Rong Jin

Self-attention is powerful in modeling long-range dependencies, but it is weak in local finer-level feature learning. The performance of local self-attention (LSA) is just on par with convolution and inferior to dynamic filters, which puzzles researchers on whether to use LSA or its counterparts, which one is better, and what makes LSA mediocre. To clarify these, we comprehensively investigate LSA and its counterparts from two sides: \emph{channel setting} and \emph{spatial processing}. We find that the devil lies in the generation and application of spatial attention, where relative position embeddings and the neighboring filter application are key factors. Based on these findings, we propose the enhanced local self-attention (ELSA) with Hadamard attention and the ghost head. Hadamard attention introduces the Hadamard product to efficiently generate attention in the neighboring case, while maintaining the high-order mapping. The ghost head combines attention maps with static matrices to increase channel capacity. Experiments demonstrate the effectiveness of ELSA. Without architecture / hyperparameter modification, drop-in replacing LSA with ELSA boosts Swin Transformer \cite{swin} by up to +1.4 on top-1 accuracy. ELSA also consistently benefits VOLO \cite{volo} from D1 to D5, where ELSA-VOLO-D5 achieves 87.2 on the ImageNet-1K without extra training images. In addition, we evaluate ELSA in downstream tasks. ELSA significantly improves the baseline by up to +1.9 box Ap / +1.3 mask Ap on the COCO, and by up to +1.9 mIoU on the ADE20K. Code is available at \url{https://github.com/damo-cv/ELSA}.

PaperPDFCodeCode Syntology ran

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

For agents, Syntology's MCP tool lists every function and class Syntology harvested from this paper and whether it ran (how to connect): get_harvested_code_for_paper(arxiv_id="2112.12786")

Code

Syntology Ran 1 of 1 code samples harvested from 1 repository linked to this paper; 0 have no recorded run. Of those that ran: 1 ran · fixture could not drive it.

By repository: official repository: 1 sample from 1 repository, 1 ran. The run record, sample by sample. “Ran” means executed on a synthesized input, not that the code is correct or reproduces the paper.

damo-cv/elsa officialmentioned in papermentioned on GitHubpytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

1 sample harvested; 1 ran; 0 honoured the contract we drafted; 0 have no recorded run. Read from Syntology's graph 2026-09-24; that is when this build read the record, not when the samples ran.

1ran · fixture could not drive it

Licence: 0 of the 1 sample are pointer only, meaning Syntology does not serve that copy's text. This page shows no code text for any sample; each one links to its file in the repository.

Harvested from damo-cv/elsa. “Ran” means the sample executed on a synthesized input. It does not mean the output is correct, and nothing here reproduces the paper's results. “Honoured” and “violated” refer to a contract Syntology drafted from the code itself; “our draft was wrong” and “fixture could not drive it” are failures of Syntology's instrument, not of the code.

Each sample ends with its code_sha256, Syntology's identity for that exact code. An agent fetches the stored sample with Syntology's MCP tool get_code(code_sha256="…") (how to connect); click an identity to copy that call.

Repository labels, per sample. official repository: The archive marks this repository official for the paper. named in the paper: The archive records that the paper mentions this repository; it is not marked official. community (archive-listed): In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper. found in paper text by Syntology: Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted. community: Not in the archive's code links for this paper; a community repository Syntology harvested. Samples from a repository marked official are listed first. Licence labels name the repository's licence as recorded at harvest. “Pointer only” means Syntology does not serve that copy's text, for one of four reasons: no licence file was found; the licence was not identified; the licence is recorded as permissive but that copy's record is not marked cleared; or the licence is outside the permissive list Syntology serves text under (MIT, Apache-2.0, BSD and similar). Some licences outside that list permit redistribution, such as WTFPL, and GPL-3.0 under its conditions; they are simply not on the list. Hover a licence label for the reason. File links open the file on GitHub at the default branch, which may have changed since the harvest.

elsa_op damo-cv/elsa/cls/models/elsa/elsa.py official repository ran · fixture could not drive it MIT (permissive) · dda6a388725848ff · report

Tasks

Image ClassificationInstance SegmentationObject DetectionSemantic Segmentation

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Image Classification ImageNet ELSA-VOLO-D5 (512*512) GFLOPs 437 #100 of 1060 Archive leaderboard report
Image Classification ImageNet ELSA-VOLO-D5 (512*512) Number of params 298M #100 of 1060 Archive leaderboard report
Image Classification ImageNet ELSA-VOLO-D5 (512*512) Top 1 Accuracy 87.2% #100 of 1060 Archive leaderboard report
Image Classification ImageNet ELSA-VOLO-D1 GFLOPs 8 #294 of 1060 Archive leaderboard report
Image Classification ImageNet ELSA-VOLO-D1 Number of params 27M #294 of 1060 Archive leaderboard report
Image Classification ImageNet ELSA-VOLO-D1 Top 1 Accuracy 84.7% #294 of 1060 Archive leaderboard report
Image Classification ImageNet ELSA-Swin-T GFLOPs 4.8 #513 of 1060 Archive leaderboard report
Image Classification ImageNet ELSA-Swin-T Number of params 28M #513 of 1060 Archive leaderboard report
Image Classification ImageNet ELSA-Swin-T Top 1 Accuracy 82.7% #513 of 1060 Archive leaderboard report
Instance Segmentation COCO minival ELSA-S (Cascade Mask RCNN) AP50 67.8 #54 of 93 Archive leaderboard report
Instance Segmentation COCO minival ELSA-S (Cascade Mask RCNN) AP75 47.8 #54 of 93 Archive leaderboard report
Instance Segmentation COCO minival ELSA-S (Cascade Mask RCNN) mask AP 44.4 #54 of 93 Archive leaderboard report
Instance Segmentation COCO minival ELSA-S (Mask RCNN) AP50 67.3 #62 of 93 Archive leaderboard report
Instance Segmentation COCO minival ELSA-S (Mask RCNN) AP75 46.4 #62 of 93 Archive leaderboard report
Instance Segmentation COCO minival ELSA-S (Mask RCNN) mask AP 43.0 #62 of 93 Archive leaderboard report
Object Detection COCO minival ELSA-S (Cascade Mask RCNN) AP50 70.5 #71 of 220 Archive leaderboard report
Object Detection COCO minival ELSA-S (Cascade Mask RCNN) AP75 56.0 #71 of 220 Archive leaderboard report
Object Detection COCO minival ELSA-S (Cascade Mask RCNN) box AP 51.6 #71 of 220 Archive leaderboard report
Object Detection COCO minival ELSA-S (Mask RCNN) AP50 70.4 #91 of 220 Archive leaderboard report
Object Detection COCO minival ELSA-S (Mask RCNN) AP75 52.9 #91 of 220 Archive leaderboard report
Object Detection COCO minival ELSA-S (Mask RCNN) box AP 48.3 #91 of 220 Archive leaderboard report
Semantic Segmentation ADE20K ELSA-Swin-S Validation mIoU 50.3 #114 of 235 Archive leaderboard report
Semantic Segmentation ADE20K val ELSA-Swin-S mIoU 50.3 #50 of 95 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Absolute Position EncodingsAdamAttentionBPEConvolutionDense ConnectionsDropoutLabel SmoothingLayer NormalizationLinear LayerMulti-Head AttentionPosition-Wise Feed-Forward LayerResidual ConnectionSoftmaxStochastic DepthSwin TransformerTransformer

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections