Lite-HRNet: A Lightweight High-Resolution Network

By Changqian Yu, Bin Xiao, Changxin Gao, Lu Yuan, Lei Zhang, Nong Sang, Jingdong Wang. Computer Vision and Pattern Recognition, 2021

426 citations recorded by the source corpus.

We present an efficient high-resolution network, Lite-HRNet, for human pose estimation. We start by simply applying the efficient shuffle block in ShuffleNet to HRNet (high-resolution network), yielding stronger performance over popular lightweight networks, such as MobileNet, ShuffleNet, and Small HRNet. We find that the heavily-used pointwise (1 × 1) convolutions in shuffle blocks become the computational bottleneck. We introduce a lightweight unit, conditional channel weighting, to replace costly pointwise (1 × 1) convolutions in shuffle blocks. The complexity of channel weighting is linear w.r.t the number of channels and lower than the quadratic time complexity for pointwise convolutions. Our solution learns the weights from all the channels and over multiple resolutions that are readily available in the parallel branches in HRNet. It uses the weights as the bridge to exchange information across channels and resolutions, compensating the role played by the pointwise (1 × 1) convolution. Lite-HRNet demonstrates superior results on human pose estimation over popular lightweight networks. Moreover, Lite-HRNet can be easily applied to semantic segmentation task in the same lightweight manner. The code and models have been publicly available at https://github.com/HRNet/Lite-HRNet.

Subfields: Artificial Intelligence, Biomedical Engineering, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/db848dda8e07d76b5fe006a79d909026584c44b1

How researchers see this paper

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

In Artificial Intelligence

1. Jingdong Wang
2. Chunhua Shen
3. Yuhui Yuan
4. Changqian Yu
5. Changxin Gao
6. Nong Sang
7. Jingbo Wang
8. Gang Yu

In Biomedical Engineering

1. Gang Yu
2. Yuhui Yuan
3. Chen Chen
4. Chunhua Shen
5. Mubarak Shah
6. Tao Jiang
7. Jingdong Wang
8. Nasser Kehtarnavaz

In Computer Vision and Pattern Recognition

1. Chunhua Shen
2. Jingdong Wang
3. Luc Van Gool
4. Dacheng Tao
5. Yi Yang
6. Xiaogang Wang
7. Gang Yu
8. Ming-Hsuan Yang

Citations in the corpus

• Deep Learning-based Human Pose Estimation: A Survey (2020, background): “[275] introduced a light-weighted HRNet named Lite-HRNet, which designed conditional channel weighting blocks to exchange information between channels and resolutions.”
• BiSeNet V2: Bilateral Network with Guided Aggregation for Real-Time Semantic Segmentation (2020, methodology): “2020) and Lite-HRNet (Yu et al. 2021) adopts multibranches to maintain high resolution.”
• Changer: Feature Interaction is What You Need for Change Detection (2022, background): “Feature fusion is a fundamental process in many DL tasks, such as in classification [33, 43], segmentation [19, 29], multimodal tasks [37], etc .”
• Review of Image Classification Algorithms Based on Convolutional Neural Networks (2021, methodology): “Later, in 2021, Lite-HRNet [176] applies efficient shuffle blocks [166,167] to HRNet [175].”
• Conditional DETR for Fast Training Convergence (2021, methodology): “SENet [14], GENet [13] abd Lite-HRNet [45] learn from the input the channel-wise weights.”
• Multi-Scale High-Resolution Vision Transformer for Semantic Segmentation (2021, methodology): “To enhance cross-resolution interaction, we insert repeated cross-resolution fusion layers at the beginning of each module following the approach in HRNet [27,35].”
• RTMPose: Real-Time Multi-Person Pose Estimation based on MMPose (2023, methodology): “Recent attempts narrow the gap with efficient network architecture [3, 56, 63] and detection-free paradigms [19,29,51], which is yet inadequate to reach satisfactory performance for industrial applications.”