top of page

2019

Xinjing Chen, Peng Wang­, Ruigang Yang, Learning Depth with Convolutional Spatial Propagation Network, Submitted to TPAMI 

 

Chenxu Luo*, Zhenheng Yang*, Peng Wang*, Yang Wang, Wei Xu, Ram Nevatia, Alan Yuille, Every Pixel Counts ++: Joint Learning of Geometry and Motion with 3D Holistic Understanding, Submitted to TPAMI (*Equal contribution) 

 

Xinyu Huang*, Peng Wang*, Xinjing Cheng, Dingfu Zhou, Qichuan Geng, Ruigang Yang, The ApolloScape Open Dataset for Autonomous Driving and its Application, Submitted to TPAMI (*Equal contribution).

Yang Wang, Peng Wang, Zhenheng Yang,  Chenxu Luo, Yi Yang, Wei Xu, UnOS: Unified Unsupervised Optical-flow and Stereo-depth Estimation by Watching Videos, CVPR 2019

Xibin Song, Peng Wang, Dingfu Zhou, Rui Zhu, Chenye Guan, Yunchao Dai, Hao Su, Hongdong Li, Ruigang Yang, ApolloCar3D: A Large 3D Car Instance Understanding Benchmark for Autonomous Driving, CVPR 2019

2018
Zhenheng Yang, Peng Wang, Yang Wang, Wei Xu, Ramakant Nevatia, Every Pixel Counts: Unsupervised Geometry Learning with Holistic 3D Motion Understanding,  ECCV Workshop of Autonomous Driving, 2018
We jointly learning edge, depth, normal, moving object mask in a self-supervised manner using video only. 
Yuhang Song, Chao Yang, Yeji Shen, Peng Wang, Qin Huang, C.-C. Jay Kuo, SPG-Net: Segmentation Prediction and Guidance Network for Image Inpainting, BMVC 2018 
Xinyu Huang, Xinjing Cheng, Qichuan Geng; Binbin Cao, Dingfu Zhou, Peng Wang, Yuanqing Lin, Yang Ruigang, The ApolloScape Dataset for Autonomous Driving, CVPRW 2018 (pdf, project page, github)
Next generation of visual based self-driving dataset, millions of densely labelled semantic video, object instance, camera pose, lanemarks, 3D car instances.  Welcome to attend our challenges 
Xinjing Chen*, Peng Wang*, Ruigang Yang, Depth Estimation via Affinity Learned with Convolutional Spatial Propagation Network,  ECCV 2018 (pdf)(code)
Learning affinity to reach fine detailed recovery of depth. No.1 in kitti stereo estimation. 
Peng Wang, Ruigang Yang, Binbin, Cao, Wei Xu, Yuanqing Lin, DeLS-3D: Deep Localization and Segmentation with a 3D Semantic Map,  CVPR 2018
A large data with dense 3D points, 3D pose tracks and semantic segments is joint release with this paper.
Zhenheng Yang, Peng Wang, Yang Wang, Wei Xu, Ramakant Nevatia, LEGO: Learning Edge with Geometry all at Once by Watching Videos,  CVPR 2018 (Spotlight Oral)
We jointly learning edge, depth and normal in a self-supervised manner using video only. 
Hao Zhu, Hao Su, Peng Wang, Ruigang Yang, View Extrapolation of Human Body from a Single Image, CVPR 2018
View synthesis on human with various shapes, which is much harder than chair or car. We alleviate the difficulties by depth and camera models.
Yang Wang, Yi Yang, Zhenheng Yang, Peng Wang, Liang Zhao, Wei Xu, Occlusion Aware Unsupervised Learning of Optical Flow, CVPR 2018
Model occlusion in optical flow estimation through discovering overlapping  after warping with flow..
Liang-Chieh  Chen, Alexander  Hermans, George  Papandreou, Florian  Schroff, Peng Wang, Hartwig  Adam, MaskLab: Instance Segmentation by Refining Object Detection with Semantic and Direction Features, CVPR 2018 (pdf)
Zhenheng Yang, Peng Wang, Wei Xu, Liang Zhao, Ramakant Nevatia,  Unsupervised Geometry Estimation with Edge-aware Depth-Normal Consistency. AAAI 2018 (Oral), New Orleans, USA
2017
Fangting Xia, Peng Wang, Alan Yuille, Joint Multi-Person Pose Estimation and Semantic Part Segmentation in a Single
Image. CVPR 2017, Honolulu, Hawaii, USA (pdf, pascal-pose-dataset)
We combined human part segmentation and pose estimation with deep learning to improve both of the tasks over the popular PASCAL Dataset.
2016
Peng Wang, Xiaohui Shen, Bryan Russel, Scott Cohen, Brian Price, Alan Yuille, SURGE: Surface Regularized Geometric Estimation from a Single Image,  NIPS 2016, Barcelona, Spain. (pdf, supplimentary, video)
We regularize depth and normal prediction by using edge and 3D surface prediction from a single image, which enforces planar equation inside a DCRF layer, and achieves better results visually and quantitatively.
Fangting Xia , Peng Wang, Liang-Chieh Chen,  Alan Yuille,  Zoom Better to See Clearer: Human and Object Part Segmentation with Auto Zoom Net, ECCV 2016, Amsterdam, Netherland (pdf)
By inducing a scale estimator for each object and it corresponding parts, we obtain much better results in terms of segmentation accuracy.

Peng Wang,  Alan Yuille, DOC: Deep OCclusion Recovering From A Single Image,  ECCV 2016 Amsterdam, Netherland. (pdf, data & toolbox)
Using the deep fully convolutional network for occlusion boundary inference from a single image, which is order of magnitude faster and more accurate than previous works. Also we labelled occlusion relationship over PASCAL 20 objects.

Fangting Xia, Peng Wang * , Jun Zhu * ,  Alan Yuille, Pose-Guided Human Parsing with Deep Learned Features, AAAI, 2016, Phoenix, USA (* indicates equal contribution) (pdf)
2015
Peng Wang, Xiaohui Shen, Zhe Lin, Scott Cohen, Brian Price, Alan Yuille, Joint Object and Part Segmentation using Deep Learned Potentials,  ICCV 2015, Chile (pdf, supplimentaryhorse_cow_data,  Pascal_animal_trainval_list , Part Challenge)
Part gives more details while object automatically provide long-range context,  a supervised part based segmentation strategy for animals. 

 

 

Peng Wang, Xiaohui Shen, Zhe Lin, Scott Cohen, Brian Price, Alan Yuille, Towards Unified Depth and Semantic Prediction from a Single Image,  CVPR 2015, Boston, USA (pdf, supplimentaryapplication video, depth results on NYU v2 test)
A fresh new joint learning and inference method for joint segmentation and depth estimation with a unified CNN & HCRF. 
 
Peng Wang, Zhe Lin,  Radomir Mech, Learning a Photo Cropping Cascade,  WACV, 2015, Hawaii, USA (pdf, supplimentary, dataset (crop label),  images)
Propose photo cropping candidates to increase visual satisfaction of your photos. An efficient cascade branch and bound algorithm for selecting sliding windows that optimizing cascade type energy or models.   Section Best Paper Award. 

 

 

Peng Wang, Alan Yuille, Error Factor Analysis for Wild-scene Image Labelling.  WACV, 2015, Hawaii, USA (pdf, code with need materials)
Better way to evaluate your segmentation beyond IOU, know where the error mostly happened.
 
2013



Peng Wang, Jingdong Wang, Gang Zeng, Weiwei Xu, Hongbin Zha, Shipeng Li, Supervised Kernel Descriptor for Visual Recognition. CVPR, 2013, Portland, USA (pdf)
A supervised approach for learning low level patch descriptors from image label for image recognition (Early version of deep representation learning,  but with bag of words framework).
2012​

Peng Wang, Jingdong Wang, Gang Zeng, Jie Feng, Hongbin Zha, Shipeng Li. Salient Object Detection for Searched Web Images via Global Saliency (Bounding box regression for Salient Object Detection). CVPR, 2012, Rhode Island, USA.(pdfproject page and dataset)
This is early work using box regression for object detection ( Related with recent direct box regression strategies with Deep Learning for object detection (e.g. Deep Salient Object, Dense box). This method handle web scale data based on random forest regression

Peng Wang,​ Dongqing Zhang, Gang Zeng, Jingdong Wang. Contextual Dominant Color Name Extraction for Web Image Search. Workshop of IEEE International Conference on Multimedia & Expo (ICME), 2012, Melbourne, Australia (pdf).
Extract human perceptually dominant color of images, considering the contextual pixels
Peng Wang, Dongqing Zhang, Jingdong Wang, Zhong Wu, Xian-Sheng Hua, Shipeng Li. Color Filter for Image Search ACM Multi Media(MM) Demo Abstract, 2012, Nara, Japan (pdfProject Page)

Peng Wang, Gang Zeng, Rui Gan, Jingdong Wang, Hongbin Zha. Structure-sensitive Superpixels via  Geodesic Distance. International Journal of Computer Vision (IJCV), 2013 (pdf). 
2011

Gang Zeng, Peng Wang, Jingdong Wang, Rui Gan, Hongbin Zha. Structure-sensitive Superpixels via  Geodesic Distance. The Thirteenth IEEE International Conference on Computer Vision (ICCV), 2011, Barcelona, Spain(pdf)
An image over-segmentation method aware of image structure information. It combines geometric flow and soft clustering



 

bottom of page