2026arXiv · Preprint
RASLF: Representation-Aware State Space Model for Light Field Super-Resolution
Representation-aware state-space modeling with geometric refinement and feature aggregation for light-field reconstruction.
Kai Jin
Staff Engineer II · BIGO · Beijing
I am a Staff Engineer II at BIGO in Beijing, working on computer vision and interactive AI applications. I connect research ideas with practical systems and user experiences.
My research spans light-field and stereo super-resolution, visual representations, and 3D scene understanding, with an emphasis on efficient visual models. I explore convolutional, Transformer and state-space architectures for spatial-angular modeling.
Outside work, I build personal assistants, native apps and games with AI-assisted coding. I also enjoy snowboarding and hiking.
News
- Jun 2026
- Our team won all three tracks of the NTIRE 2026 Light Field Image Super-Resolution Challenge: Classical, Efficiency and Large Model.
- Mar 2026
- RASLF was released as a preprint on representation-aware state-space modeling for light-field super-resolution.
- Jun 2025
- LFTransMamba appeared at CVPR Workshops; our team placed 1st, 2nd and 1st across the three NTIRE 2025 light-field SR tracks.
- Jun 2023
- DistgEPIT appeared at CVPR Workshops and won the NTIRE 2023 Light Field Image Super-Resolution Challenge.
Selected Publications
2025CVPR Workshops1st / 2nd / 1st · NTIRE 2025
LFTransMamba: A Hybrid Mamba-Transformer Model for Light Field Image Super-Resolution
Combines Transformer context modeling, Mamba and enhanced position-sensitive windowing for light-field super-resolution.
2025IEEE Transactions on Computational Imaging
L²FMamba: Lightweight Light Field Image Super-Resolution with State Space Model
Efficient spatial-angular modeling with state-space networks for lightweight light-field super-resolution.
2023CVPR WorkshopsWinner · NTIRE 2023
DistgEPIT: Enhanced Disparity Learning for Light Field Image Super-Resolution
Hybrid convolutional and Transformer modules learn disparity correlations, with position-sensitive windowing for consistent inference.
2022CVPR WorkshopsRunner-up · NTIRE 2022
SwiniPASSR: Swin Transformer Based Parallax Attention Network for Stereo Image Super-Resolution
Swin Transformer and bidirectional parallax attention exploit cross-view information for stereo image super-resolution.
* Equal contribution. Selected works only; figures are from the corresponding papers.
Selected Patents
Selected inventions listing Kai Jin as an inventor. Related CN and WO publications are grouped together.
Granted patents
- Target-region image detail enhancement
CN116596774B · Grant published 3 Jul 2026 · Co-inventor · BIGO Technology Pte Ltd.
- No-reference video quality determination
CN114596259B · Grant published 25 Jul 2025 · Co-inventor · BIGO Technology Pte Ltd. · Related PCT publication: WO2023138590A1.
- Road-image recognition model construction and image recognition
CN111507226B · Grant published 11 Aug 2023 · Co-inventor · Beijing Juefei Technology Co., Ltd.
Published applications
- Video quality evaluation
CN116471262A · Application published 21 Jul 2023 · Co-inventor · BIGO Technology Pte Ltd.
- 3D face reconstruction
CN115330947A · Application published 11 Nov 2022 · Inventor · BIGO Technology Pte Ltd. · Related PCT publication: WO2024032464A1.
- Inference-task processing
CN111401560A · Application published 10 Jul 2020 · Co-inventor · Beijing Juefei Technology Co., Ltd.
Publication records checked 28 Oct 2026. WO numbers identify international application publications.
Honors & Awards
NTIRE Light Field Image Super-Resolution Challenge
CVPR Workshop · Denver
Winner · ClassicalWinner · EfficiencyWinner · Large Model
NTIRE Light Field Image Super-Resolution Challenge
CVPR Workshop · Nashville
Winner · ClassicalRunner-up · EfficiencyWinner · Large Model
NTIRE Light Field Image Super-Resolution Challenge
CVPR Workshop · Vancouver
Winner
Urban3D Challenge on Urban Scenes Understanding
ECCV Workshop · 2nd challenge
Winner · Instance SegmentationRunner-up · Semantic Segmentation
NTIRE Stereo Image Super-Resolution Challenge
CVPR Workshop · New Orleans
Runner-up
Urban3D Challenge on Urban Scenes Understanding
ICCV Workshop · 1st challenge
Winner
Team awards, listed by year and track. Certificates retain the original author lists.
Contact
I welcome conversations about computer vision, applied AI and ideas worth building.
Email: jinkai@bigo.sg