- Generation and Analysis of
   Pixels, Points and Polygons -
Deep Fashion3D [Download link, Project Page]
(ECCV 2020 Oral)
- It contains 2078 colored models, in point cloud format, reconstructed from real garments, which covers 10 different categories and 563 garment instances.
3DCaricShop [Download link, Project Page]
(CVPR 2021)
- It contains 2000 high-quality diversified 3D caricatures with a paired 2D image. All the 3D models are manually crafted by professional artists. 3DCaricShop also provides rich annotations including camera parameters and 3D facial landmarks.
SUNREFER [Download link, Project Page]
(CVPR 2021)
- It is a large-scale dataset of referring expression dedicated for visual language research in RGBD images. It contains 38,495 annotations of referring expression on 7,699 RGBD images.
DeepSketch2Face [Download link]
- Our sketching system allows amateur users to create 3D face or caricature models with complicated shape and expression in a few minutes.
OCPose [Download link, Project Page]
(ECCV 2020)
- It 9,000 images with 18,000 person instances, all joints of each individual is manually labeled. It contains challenging invisible joints and complex intertwined human poses.
AniHead [Download link]
(IEEE TVCG 2020)
- It contains 220 high-quality 3D meshes of Animal heads, in 19 categories.
(CVPR 2021) [Github]

- Given a single RGB-D image of an indoor scene and a query language text. Our algorithm can automatically output the 3D bounding box of the most relevant object.
(CVPR 2021) [Github]

- Given a 3D scene in point cloud format, our algorithm can output the 3D bounding boxes of detected objects and also there reconstructed meshes.
(NeurIPS 2020) [Github]

- Given a partial scan of an object, our method can recover the missing regions and output the completed surface of the object.
(CVPR 2020 Oral) [Github]

- Given a singe image of an indoor scene, our method can simutaneously recover the 3D room layout, all 3D bounding boxes of detected objects and also the reconstructed meshes of the objects.
(ECCV 2020) [Github]

- Given a singe image of multiple person individuals, our method can output the pose joints of all individuals. 
(CVPR 2020) [Github]

- A novel convolution operator for point cloud, it can support 3D shape analysis and 3D scene understanding, such as object classification and scene semantic segmentation.
(ACMMM 2020) [Github]

- Given a video where a person is dancing, we ask a user to imitate some key frames and provide several source images. Our method can generate a video with motion that is consistent with the target person but with appearance from the user.
HEMLets Posh
(ICCV 2019, TPAMI 2021) [Github]

- Given a single human image, our method firstly detects person individuals and then outputs their 3D pose and shapes.
(ICCV 2019) [Github]

- Given a single-view image of an object (with a segmented mask), our method can automatically output the 3D mesh of the object.
Skeleton-bridged SVR
(CVPR 2019 Oral) [Github]

- Given a single-view image of an object (cutted out from its background), our method can automatically output the 3D mesh of the object.
(SIGGRAPH 2016) [Github]

- Given a triangle mesh and a source vertex, our method can calculate the exact geodesics from the source to all the other vertices.
(SIGGRAPH 2015) [Github]

- An L1-Flattening which can work on edge-preserving smoothing and also support intrinsic image decomposion.