Recent Posts

ProSR won the Winner Award for NTIRE2018 track 1, and the Honorable Mention Award for track 2~3. Congratuations to all other winning teams.


ProSR took the 2nd place in NTIRE 2018 Super-Resolution Challenge. Our paper is accepted to the NTIRE 2018 workshop and I’ll be giving a 10-min talk during the workshop.


Selected Publications

We present a detail-driven deep neural network for point set upsampling. A high-resolution point set is essential for point-based rendering and surface reconstruction. Inspired by the recent success of neural image super-resolution techniques, we progressively train a cascade of patch-based upsampling networks on different levels of detail end-to-end. We propose a series of architectural design contributions that lead to a substantial performance boost. The effect of each technical contribution is demonstrated in an ablation study. Qualitative and quantitative experiments show that our method significantly outperforms the state-of-the-art learning-based and optimazation-based approaches, both in terms of handling low-resolution inputs and revealing high-fidelity details.
arXiv, 2018

We set a new benchmark for single-image super-resolution by exploiting progressiveness both in architecture and training. The proposed multi-scale models, ProSR and ProSRGan, improve the reconstruction quality in terms of PSNR and visual quality respectively. ProSR is one of the winning teams.
CVPRW 2018, 2018

We propose a new deep architecture by incorporating object/human detection results into the framework for action recognition, called two-stream semantic region based CNNs (SR-CNNs). We perform experiments on UCF101 dataset and demonstrate its superior performance to the original two-stream CNNs.
BMVC, 2016

Recent Publications

Patch-base progressive 3D Point Set Upsampling

Preprint PDF Code Dataset

A Fully Progressive Approach to Single-Image Super-Resolution

Preprint Supplement Code Video Download

Two-Stream SR-CNNs for Action Recognition in Videos

Preprint PDF Code Dataset