1. Medical IMage analysis
Dec. 2016 ~ Present
Devoted to develop models and frameworks for automatic medical image analysis, especially for the volumetric medical image segmentation.
Devoted to develop models and frameworks for automatic medical image analysis, especially for the volumetric medical image segmentation.
2. OBJECT RECOGNITION
Apr. 2016 ~ Sep. 2016
We train deep neural networks on the foreground (object) and background (context) regions of images respectively. Considering human recognition in the same situations, networks trained on pure background without objects achieves highly reasonable recognition performance that beats humans to a large margin if only given context. However, humans still outperform networks with pure ob- ject available, which indicates networks and human beings have different mechanisms in understanding an image. arXiv:1611.06596. |
3. Multiple Instance learning
Jan. 2015 ~ Jul. 2015
We proposed a novel formulation for Multiple Instance Learning(MIL), which solves an optimazation problem by considering both the bag-level loss and instance-level loss for the multi-instance problems. The approach achieved promising results on several standard MIL datasets including Fox, Elephant and Tiger datasets. Our submitted paper was accepted by the ICCV 2015. arXiv:1510:01027. |
4. 3D Shape Retrieval
Apr. 2014 ~ Jan. 2015
1. Zhuotun Zhu, Xinggang Wang, Song Bai, Cong Yao and Xiang Bai. Deep Learning Representation using Autoencoder for 3D Shape Retrieval. ICSPAC, 2014, Wuhan. arXiv:1409.7164 [cs.CV]. The extension of this conference paper was accepted to the special issue of Neurocomputing. 2. Adopting Two Layer Coding framework, we achieved the state-of-the-art performance on five datasets. In April. 2015, the paper was accepted in the Transactions on Pattern Analysis and Machine Intelligence (TPAMI). |
5. vehicle recognition
Jul. 2014 ~ Aug. 2014
The discriminative patterns are image patches that can tell the intrinsic differences between different car brand categories. The goal is to discover and make use of such discriminative patterns for the task of car brand recognition. In this framework, a novel Multiple Instance Learning method is proposed. |