HRFormer: High-Resolution Vision Transformer for Dense Predict

By Yuhui Yuan, Rao Fu, Lang Huang, Weihong Lin, Chao Zhang, Xilin Chen, Jingdong Wang. 2021

342 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Model Architectures

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

In Model Architectures

Citations in the corpus

SimpleClick: Interactive Image Segmentation with Simple Vision Transformers (2022, neutral): “Recently, ViT-based approaches [17, 24, 44, 46, 50] have shown competitive performance on segmentation tasks compared to ConvNets.”
SegNeXt: Rethinking Convolutional Attention Design for Semantic Segmentation (2022, methodology): “SegFormer [80] HRFormer [88] MaskFormer [11] SegNeXt Figure 1: Performance-Computing curves on the Cityscapes (left) and ADE20K (right) validation sets.”
Image De-Raining Transformer (2022, background): “Based on the original FFN which consists of two point-wise MLPs and a non-linear activation in between, LeFF [90], [95] adds a 3 (cid:4) 3 depth-wise convolution and non-linear activation between two MLPs to enhance locality.”
Neural operators for accelerating scientific simulations and design (2023, neutral): “Even if modifications are made to incorporate multiple resolutions at inputs or outputs [14], they are not principled and have no guarantees for capturing finer scales.”
EfficientViT: Lightweight Multi-Scale Attention for High-Resolution Dense Prediction (2023, neutral): “SegFormer [51] HRFormer [55] SegNeXt [18] EfficientViT We summarize the comparison between our work and prior SOTA high-resolution dense prediction models in Table 1.”
Activating More Pixels in Image Super-Resolution Transformer (2022, background): “Although vision Transformer has shown its superiority on modeling long-range dependency [14,45], there are still many works demonstrating that the convolution can help Transformer achieve better visual representation [26,59,61,63,64].”
Deep Learning-based Human Pose Estimation: A Survey (2020, background): “Recently, due to the superior performance, the HRNet [222] and its variations [32, 275, 278] have been widely adopted in HPE and other pose-related tasks.”
HRDA: Context-Aware High-Resolution Domain-Adaptive Semantic Segmentation (2022, background): “This concept was further improved in different aspects including increasing the receptive field while preserving spatial details [3,4,7,52,68,78,88], integrating context information [31,79,81,83,96], attention mechanisms [20,33,71,89], refining boundaries [17,37,82], neural architecture search [2, 4”
iSegFormer: Interactive Segmentation via Transformers with Application to 3D Knee MR Images (2021, methodology): “To further demonstrate the efﬁ-ciency of iSegFormer , we also implemented iSegFormer using two recently proposed Transformer backbones for segmentation: HRFormer [11] and SegFormer [12].”