Yancong Lin

Yancong Lin (林彦聪)

Research on built-in knowledge priors for scalable perception models.
Collaboration with AIIR on automated inspection for aircraft engines.

Email / Google Scholar / Github / LinkedIn /

About me

I am interested in creating data-efficient AI models for various computer vision/robotics perception tasks by pre-wiring deep learning with generic innate knowledge priors. This eliminates the need to collect and annotate large datasets, as built-in knowledge no longer has to be learned. My work exhibits leading performance while reducing data demand substantially, particularly in processing images and point clouds.

News

27/02/2025 Our submission to CVPR'25 on motion-rigid scene flow estimation has been accepted.

21/01/2025 Master thesis: Congrats to Lucas on a nice thesis - Mutation Delineation of Maps Using Aerial Imagery.

01/10/2024 Moved to ETH Zurich as an academic guest at the Institute of Geodesy and Photogrammetry.

17/07/2024 Master defense: Arbër, SfM applications for 3D reconstruction from 2D avionics industrial inspection videos.

11/07/2024 Master defense: Tibbe, MultiViT: 2D to 3D transfer learning using a jointly optimized Vision Transformer.

17/06/2024 ICP-Flow has won the Argoverse 2 Scene Flow Challenge (unsupervised track).

28/02/2024 Our submission to CVPR'24 on LiDAR scene flow estimation has been accepted.

29/01/2024 Our submission to ICRA'24 on active learning for LiDAR semantic segmentation has been accepted.

18/12/2023 Research grant:NWO NGF AiNed XS Europa (80k euros), in collaboration with Prof. Konrad Schindler, ETH.

13/07/2023 Our submission to ICCV'23 on why classification helps regression has been accepted.

26/06/2023 Master defense: Nafie, BladeNeRF. Cum laude!

19/10/2022 I am an outstanding reviewer, ECCV'22.

01/10/2022 Postdoc at the Intelligent Vehicles Group, 3mE, TUDelft.

30/09/2022 Our submission to BMVC'22 on mirror symmetries has been accepted.

21/05/2022 I am an outstanding reviewer, CVPR'22.

25/04/2022 I have successfully defended my phd dissertation.

03/03/2022 Our submission to CVPR'22 on vanishing points has been accepted.

01/01/2022 New project with an industrial partner: computer vision for aircraft engine inspection.

18/08/2021 Our submission to ICCV'21 workshop (Deep Learning for Geometric Computing): Best student paper award.

23/07/2021 ICCV'21 workshop: 2nd Visual Inductive Priors for Data-Efficient Deep Learning Workshop,

12/07/2021 Master: Andrea, Investigating transformers in the decomposition of polygonal shapes as point collections.

26/05/2021 Our submission to ICIP'21 has been accepted.

01/10/2020 Master defense: Kang Lang, Vertex-voting-based polygonal object detection.

23/08/2020 ECCV'20 workshop: 1st Visual Inductive Priors for Data-Efficient Deep Learning Workshop,

04/07/2020 Our submission to ECCV'20 on wireframe parsing has been accepted.

Publications
	VoteFlow: Enforcing Local Rigidity in Self-Supervised Scene Flow Yancong Lin , Shiming Wang (co-first author), LiangLiang Nan, Julian Kooij, Holger Caesar CVPR, 2025 arXiv / Code feature learning + translation voting for identifying shared motion from adjacent LiDAR scans.
	Bosch Street Dataset: A Multi-Modal Dataset with Imaging Radar for Automated Driving ..., Yancong Lin , ... ArXiv, 2024 arXiv BSD offers a unique integration of high-resolution imaging radar, lidar, and camera sensors.
	ICP-Flow: LiDAR Scene Flow Estimation with ICP Yancong Lin , Holger Caesar CVPR, 2024 arXiv / Code Clustering + ICP, w/o learning, w/o training data: a strong baseline for LiDAR scene flow. Winner, (Unsupervised) Argoverse 2 Scene Flow Challenge, CVPR 2024.
	BaSAL: Size Balanced Warm Start Active Learning for LiDAR Semantic Segmentation Jiarong Wei, Yancong Lin (corresponding author), Holger Caesar ICRA, 2024 arXiv / Code Class imbalance and cold start in active learning: size-based partition to rescue
	A step towards understanding why classification helps regression Silvia-Laura Pintea, Yancong Lin, Jouke Dijkstra, Silvia-Laura Pintea, Jan van Gemert ICCV, 2023 arXiv / Code For a regression task, if the data sampling is imbalanced, then add a classification loss.
	Deep Vanishing Point Detection: Geometric priors make dataset variations vanish Yancong Lin, Ruben Wiserma, Silvia-Laura Pintea, Klaus Hildebrandt, Elmar Eisemann, Jan van Gemert CVPR, 2022 arXiv / Code Hough Transform and Gaussian sphere priors for data-efficient and domain-robust vanishing point detection.
	NeRD++: Improved 3D-mirror symmetry learning from a single image Yancong Lin, Silvia-Laura Pintea, Jan van Gemert BMVC, 2022 arXiv / Code Detecting 3D mirror plane from a single image, using feature correlations, mirroring, multi-scale spherical convs.
	Investigating transformers in the decomposition of polygonal shapes as point collections Andrea Alfieri, Yancong Lin, Jan van Gemert ICCV Workshop (Best student paper!), 2021 arXiv Exploit auto-regressive and parallel transformers in predicting collections of points (polygons).
	Semi-Supervised Lane Detection with Deep Hough Transform Yancong Lin, Silvia-Laura Pintea, Jan van Gemert ICIP, 2021 Code / arXiv Exploit lane representations in the Hough space from unlabel data.
	Deep Hough Transform Line Priors Yancong Lin, Silvia-Laura Pintea, Jan van Gemert ECCV, 2020 Code / arXiv Reduce data dependency by adding line priors through a trainable Hough transform module.
	Quality Index for Stereoscopic Images by Jointly Evaluating Cyclopean Amplitude and Cyclopean Phase Yancong Lin, Jiachen Yang, Wen Lu, Qinggang Meng, Zhihan Lv, Houbing Song IEEE Journal of Selected Topics in Signal Processing, 2016 Stereo image quality assessment using binocular vision models and low-level features. Core contribution of master thesis.