</> Code with Papers by Syntology

Methods › General › Attention Modules › Deformable Attention Module

Deformable Attention Module

42 papers tagged archive 2025-07-28

Introduced by Xizhou Zhu et al. in Deformable DETR: Deformable Transformers for End-to-End Object Detection

archive 2025-07-28 Description, source and code snippet are the archive's method entry.

Deformable Attention Module is an attention module used in the Deformable DETR architecture, which seeks to overcome one issue base Transformer attention in that it looks over all possible spatial locations. Inspired by deformable convolution, the deformable attention module only attends to a small set of key sampling points around a reference point, regardless of the spatial size of the feature maps. By assigning only a small fixed number of keys for each query, the issues of convergence and feature spatial resolution can be mitigated.

Given an input feature map x ∈ℝ^(C ×H ×W), let q index a query element with content feature 𝐳_q and a 2-d reference point 𝐩_q, the deformable attention feature is calculated by:

DeformAttn(𝐳_q, 𝐩_q, 𝐱)=∑ₘ₌₁ᴹ 𝐖ₘ[∑ₖ₌₁ᴷ A_(m q k) ·𝐖ₘ^' 𝐱(𝐩_q+Δ𝐩_(m q k))]

where m indexes the attention head, k indexes the sampled keys, and K is the total sampled key number (K ≪H W) . Δp_(m q k) and A_(m q k) denote the sampling offset and attention weight of the kᵗʰ sampling point in the mᵗʰ attention head, respectively. The scalar attention weight A_(m q k) lies in the range [0,1], normalized by ∑ₖ₌₁ᴷ A_(m q k)=1 . Δ𝐩_(m q k) ∈ℝ² are of 2-d real numbers with unconstrained range. As p_q+Δp_(m q k) is fractional, bilinear interpolation is applied as in Dai et al. (2017) in computing 𝐱(𝐩_q+Δ𝐩_(m q k)). Both Δ𝐩_(m q k) and A_(m q k) are obtained via linear projection over the query feature z_q . In implementation, the query feature z_q is fed to a linear projection operator of 3 M K channels, where the first 2 M K channels encode the sampling offsets Δp_(m q k), and the remaining M K channels are fed to a softmax operator to obtain the attention weights A_(m q k).

PaperSource

Papers archive 2025-07-28

30 shown of 42, newest first. Repository counts are the archive's code-links table. A Syntology line states what Syntology ran from that paper's harvested code; it is per sample and not a correctness claim.

Tasks archive 2025-07-28

20 shown of 66 tasks the archive attaches to papers tagged with this method, by distinct papers. A task without a page in the catalog is plain text.

TaskPapers
Object Detection28
object-detection23
Object17
Decoder10
Instance Segmentation4
2D Object Detection3
Autonomous Driving3
GPU3
Knowledge Distillation3
Pedestrian Detection3
Semantic Segmentation3
Language Modeling2
Language Modelling2
Object Localization2
Optical Flow Estimation2
Real-Time Object Detection2
Region Proposal2
Segmentation2
Semi-Supervised Object Detection2
Video Instance Segmentation2

Usage over time archive 2025-07-28

Papers per year tagged with Deformable Attention Module: 2020 to 2025, peak 12 12 0 2020: 1 paper 2020 2021: 7 papers 2021 2022: 12 papers 2022 2023: 11 papers 2023 2024: 10 papers 2024 2025: 1 paper 2025
Papers per year the archive tags with this method, by the paper's archive date (42 dated). Bars are counts, not a trend claim.

Components: the archive holds no method-to-method composition, so PwC's Components table cannot be rebuilt; the Papers list carries no Results column for the same reason (the archive does not join its leaderboard rows to method tags).

Categories archive 2025-07-28

Attention Modules

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