Vision Transformer with Deformable Attention

Xia, Zhuofan; Pan, Xuran; Song, Shiji; Li, Li Erran; Huang, Gao

doi:10.48550/arxiv.2201.00520

Cited by 14 publications

(24 citation statements)

References 37 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For larger EATFormer-small and EATFormer-base models, we consistently get better results than recent counterparts, which surpass Swin-T by +2.4↑/+2.1↑ and Swin-S by +1.5↑/+1.7↑ with 1× schedule, while by +1.4↑/+1.3↑ and by +0.5↑/+0.9↑ with 3× schedule. Also, we obtain slightly higher results than DAT [114] with computation amount going down by 29G↓.…”

Section: Resultsmentioning

confidence: 75%

“…Then KV is obtained with the new feature map X, i.e., KV = f kv ( X). It is worth mentioning that the main difference between MD-MSA and recent similar work [114] lies in the modulation operation, where MD-MSA could apply appropriate attention to different position features to obtain better results. Also, any form of position embedding is not used since it makes no contribution to results, and detailed comparative experiments can be viewed in Section 5.4.3.…”

Section: Re-samplementioning

confidence: 99%

See 1 more Smart Citation

EATFormer: Improving Vision Transformer Inspired by Evolutionary Algorithm

Zhang¹,

Li²,

Wang³

et al. 2022

Preprint

View full text Add to dashboard Cite

Motivated by biological evolution, this paper explains the rationality of Vision Transformer by analogy with the proven practical Evolutionary Algorithm (EA) and derives that both have consistent mathematical formulation. Then inspired by effective EA variants, we propose a novel pyramid EATFormer backbone that only contains the proposed EA-based Transformer (EAT) block, which consists of three residual parts, i.e., Multi-Scale Region Aggregation (MSRA), Global and Local Interaction (GLI), and Feed-Forward Network (FFN) modules, to model multi-scale, interactive, and individual information separately. Moreover, we design a Task-Related Head (TRH) docked with transformer backbone to complete final information fusion more flexibly and improve a Modulated Deformable MSA (MD-MSA) to dynamically model irregular locations. Massive quantitative and quantitative experiments on image classification, downstream tasks, and explanatory experiments demonstrate the effectiveness and superiority of our approach over State-Of-The-Art (SOTA) methods. E.g., our Mobile (1.8M), Tiny (6.1M), Small (24.3M), and Base (49.0M) models achieve 69.4, 78.4, 83.1, and 83.9

show abstract

Section: Resultsmentioning

confidence: 75%

Section: Re-samplementioning

confidence: 99%

EATFormer: Improving Vision Transformer Inspired by Evolutionary Algorithm

Zhang¹,

Li²,

Wang³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…But nearly none of them makes full use of the rich shortand long-range dependencies generated by the self-attention mechanism. In the field of image processing, Deformable Attention Transformer (DAT) proposed in [137] generates the deformed sampling points by introducing an offset network. It achieves consistently-improved results on comprehensive benchmarks and reduces computational costs.…”

Section: A Discussionmentioning

confidence: 99%

Transformers in 3D Point Clouds: A Survey

Lu¹,

Quan²,

Wei³

et al. 2022

Preprint

View full text Add to dashboard Cite

In recent years, Transformer models have been proven to have the remarkable ability of long-range dependencies modeling. They have achieved satisfactory results both in Natural Language Processing (NLP) and image processing. This significant achievement sparks great interest among researchers in 3D point cloud processing to apply them to various 3D tasks. Due to the inherent permutation invariance and strong global feature learning ability, 3D Transformers are well suited for point cloud processing and analysis. They have achieved competitive or even better performance compared to the state-of-the-art non-Transformer algorithms. This survey aims to provide a comprehensive overview of 3D Transformers designed for various tasks (e.g. point cloud classification, segmentation, object detection, and so on). We start by introducing the fundamental components of the general Transformer and providing a brief description of its application in 2D and 3D fields. Then, we present three different taxonomies (i.e., Transformer implementation-based taxonomy, data representation-based taxonomy, and task-based taxonomy) for method classification, which allows us to analyze involved methods from multiple perspectives. Furthermore, we also conduct an investigation of 3D self-attention mechanism variants designed for performance improvement. To demonstrate the superiority of 3D Transformers, we compare the performance of Transformer-based algorithms in terms of point cloud classification, segmentation, and object detection. Finally, we point out three potential future research directions, expecting to provide some benefit references for the development of 3D Transformers.

show abstract

“…By incorporating the pyramid structure from CNNs, PVT [38] serves as a versatile backbone for many dense prediction tasks. DeformableViT [40] equips with a deformable self-attention module in line with deformable convolution [11] to enable flexible spatial locations conditioned on input data. ViT-Slim [5] searches for a sub-transformer network across three dimensions of input tokens, MHSA and MLP modules with a 1 -regularized soft mask to indicate the global importance of dimensions, just like the CNNs network slimming [29].…”

Section: Related Workmentioning

confidence: 99%

Super Vision Transformer

Lin¹,

Chen²,

Zhang³

et al. 2022

Preprint

View full text Add to dashboard Cite

We attempt to reduce the computational costs in vision transformers (ViTs), which increase quadratically in the token number. We present a novel training paradigm that trains only one ViT model at a time, but is capable of providing improved image recognition performance with various computational costs. Here, the trained ViT model, termed super vision transformer (SuperViT), is empowered with the versatile ability to solve incoming patches of multiple sizes as well as preserve informative tokens with multiple keeping rates (the ratio of keeping tokens) to achieve good hardware efficiency for inference, given that the available hardware resources often change from time to time. Experimental results on ImageNet demonstrate that our SuperViT can considerably reduce the computational costs of ViT models with even performance increase. For example, we reduce 2× FLOPs of DeiT-S while increasing the Top-1 accuracy by 0.2% and 0.7% for 1.5× reduction. Also, our SuperViT significantly outperforms existing studies on efficient vision transformers. For example, when consuming the same amount of FLOPs, our SuperViT surpasses the recent state-of-the-art (SoTA) EViT by 1.1% when using DeiT-S as their backbones. The project of this work is made publicly available at https://github.com/lmbxmu/SuperViT.

show abstract

Vision Transformer with Deformable Attention

Cited by 14 publications

References 37 publications

EATFormer: Improving Vision Transformer Inspired by Evolutionary Algorithm

EATFormer: Improving Vision Transformer Inspired by Evolutionary Algorithm

Transformers in 3D Point Clouds: A Survey

Super Vision Transformer

Contact Info

Product

Resources

About