Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Visual Saliency Transformer

About

Existing state-of-the-art saliency detection methods heavily rely on CNN-based architectures. Alternatively, we rethink this task from a convolution-free sequence-to-sequence perspective and predict saliency by modeling long-range dependencies, which can not be achieved by convolution. Specifically, we develop a novel unified model based on a pure transformer, namely, Visual Saliency Transformer (VST), for both RGB and RGB-D salient object detection (SOD). It takes image patches as inputs and leverages the transformer to propagate global contexts among image patches. Unlike conventional architectures used in Vision Transformer (ViT), we leverage multi-level token fusion and propose a new token upsampling method under the transformer framework to get high-resolution detection results. We also develop a token-based multi-task decoder to simultaneously perform saliency and boundary detection by introducing task-related tokens and a novel patch-task-attention mechanism. Experimental results show that our model outperforms existing methods on both RGB and RGB-D SOD benchmark datasets. Most importantly, our whole framework not only provides a new perspective for the SOD field but also shows a new paradigm for transformer-based dense prediction models. Code is available at https://github.com/nnizhang/VST.

Nian Liu, Ni Zhang, Kaiyuan Wan, Ling Shao, Junwei Han• 2021

Related benchmarks

TaskDatasetResultRank
Salient Object DetectionDUTS (test)
M (MAE)0.037
368
Camouflaged Object DetectionCOD10K (test)
S-measure (S_alpha)0.781
306
Salient Object DetectionECSSD
MAE0.033
249
RGB-D Salient Object DetectionSTERE
S-measure (Sα)0.913
232
Camouflaged Object DetectionCOD10K
S-measure (S_alpha)0.781
217
Salient Object DetectionHKU-IS
MAE0.029
202
Salient Object DetectionPASCAL-S
MAE0.062
196
Salient Object DetectionPASCAL-S (test)
MAE0.06
149
Salient Object DetectionDUT-OMRON
MAE0.058
148
RGB-D Salient Object DetectionNJU2K (test)
S-measure (Sα)0.922
137
Showing 10 of 75 rows
...

Other info

Code

Follow for update