Lin GU   谷林

I am now an assistant professor at Tohoku University (東北大学).

I am focused on developing a new generation of artificial intelligence through an evolutionary approach. Existing machine learning methods are designed to mimic human data and are therefore constrained by the limitations of the human brain. To transcend these limitations, I am currently working on directly acquiring recognition abilities and motor skills similar to those developed by humans through evolution.

Following this evolutionary approach, I have developed techniques across various domains, including computer vision, medical imaging, large language models (LLMs), robotics, and even nuclear fusion. I have authored over 100 papers in journals and conferences such as Nature Methods, PAMI, IJCV, AAAI, CVPR, ECCV, ICCV, ICLR, and MICCAI. Additionally, I serve as an associate editor for Pattern Recognition, IEEE Trans of Circuits and Systems for Video Technology (TCVST)  and as a senior/area chair for ACL, CVPR, ICCV, ICML, ICLR, NeurIPS, and WACV.

News:

We have three papers accepted at CVPR 2026.

Our paper DNGaussian++: Improving Sparse-View Gaussian Radiance Fields with Depth Normalization is accepted by PAMI.

I am glad to serve as associate editor for IEEE Trans of Circuits and Systems for Video Technology (TCSVT)

We have one spotlight and one poster paper accepted at Neurips 2025

I have 3 papers listed as top 20 papers in Google 2025 Metrics: PR (7), TMM (19), BMVC (1)

Our paper Spatial Frequency Modulation for Semantic Segmentation  is accepted by PAMI.

Our paper Investigating Synthetic-to-Real Transfer Robustness for Stereo Matching and Optical Flow Estimation  is accepted by PAMI.

Our paper Frequency-Dynamic Attention Modulation For Dense Prediction  is accepted at ICCV 2025.

We have two papers accepted at CVPR 2025.

We have three papers accepted at AAAI 2025.

Cybernetic Avatar, an open access book about cutting-edge technologies for the development of Cybernetic Avatars I have contributed has been published.

Our paper TinyLUT: Tiny Look-Up Table for Efficient Image Restoration at the Edge has been accepted by Neurips 2024.

Our paper Learning from Human Attention for Attribute-assisted Visual Recognition has been accepted by PAMI.

Our paper Rethinking masked image modeling for medical image representation is published at Medical Image Analysis.

Our paper Frequency-aware Feature Fusion for Dense Image Prediction has been accepted by PAMI.

Our nuclear fusion paper Using Convolutional Neural Networks to detect Edge Localized Modes in DIII-D from Doppler Backscattering measurements has been accepted by Review of Scientific Instruments

Our paper Interpretable Medical Image Visual Question Answering via Multi-Modal Relationship Graph Learning has been accepted by Medical Image Analysis.

Our paper A New Benchmark: Clinical Uncertainty and Severity Aware Labeled Chest X-Ray Images with Multi-Relationship Graph Learning  has been accepted by IEEE Transactions on Medical Imaging 

We have three papers accepted at ECCV 2024.

Our journal paper, BigNeuron: a resource to benchmark and predict the performance of algorithms for automated tracing of neurons in light microscopy datasets, is published by Nature Methods.

RIKEN has issued a news release about our ECCV 2022 paper: Exploring Resolution and Degradation Clues as Self-supervised Signal for Low Quality Object Detection.