AiATrack: Attention in Attention for Transformer Visual Tracking

By Shenyuan Gao, Chunluan Zhou, Chao Ma, Xing Wang, Junsong Yuan. European Conference on Computer Vision, 2022

416 citations recorded by the source corpus.

Transformer trackers have achieved impressive advancements recently, where the attention mechanism plays an important role. However, the independent correlation computation in the attention mechanism could result in noisy and ambiguous attention weights, which inhibits further performance improvement. To address this issue, we propose an attention in attention (AiA) module, which enhances appropriate correlations and suppresses erroneous ones by seeking consensus among all correlation vectors. Our AiA module can be readily applied to both self-attention blocks and cross-attention blocks to facilitate feature aggregation and information propagation for visual tracking. Moreover, we propose a streamlined Transformer tracking framework, dubbed AiATrack, by introducing efficient feature reuse and target-background embeddings to make full use of temporal references. Experiments show that our tracker achieves state-of-the-art performance on six tracking benchmarks while running at a real-time speed.

Subfields: Computer Vision and Pattern Recognition, Global and Planetary Change

View Paper: https://www.semanticscholar.org/paper/827f07e79dad0c1c77c68fa69bc4f098634d3d58

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

1. Huchuan Lu
2. Dong Wang
3. Houwen Peng
4. Ming-Ming Cheng
5. Ming-Hsuan Yang
6. Luc Van Gool
7. Jiashi Feng
8. Jianbing Shen

In Global and Planetary Change

1. Huchuan Lu
2. Houwen Peng
3. Dong Wang

Citations in the corpus

• Emergent Correspondence from Image Diffusion (2023, background): “Establishing visual correspondences between different images is crucial for various computer vision tasks such as Structure-from-Motion / 3D reconstruction [2, 74, 61, 75], object tracking [23, 93], image recognition [64, 81, 9] and segmentation [51, 48, 73, 30].”
• SeqTrack: Sequence to Sequence Learning for Visual Object Tracking (2023, background): “STARK [53] and transformer-based trackers [4,10,17,44] design corner head networks to predict the bounding box corners of target objects, as visualized in Fig.”