Today, 19:30 | graphics special session - Gao Lin, teacher team of Institute of computing technology, Chinese Academy of Sciences

Click on the blue words

Pay attention to our

AI TIME Welcome to everyone AI Fans join in ！

6 month 8 Japan 、9 Japan 19:30, In this issue, we specially invite the students of Mr. Gao Lin's team from the Institute of computing technology, Chinese Academy of Sciences to bring you wonderful sharing , Bring us in “ graphics ” The world of ！

Bili Bili live channel

Sweep yards attention AI TIME Bli bli official account reservation live broadcast

6 month 8 Japan 19:30-21:00

Yang Jie :

Doctoral candidate of Institute of computing technology, Chinese Academy of Sciences , The instructors are Mr. Gao Lin and Mr. Xia Shihong . His research interests are geometry processing and geometry learning , Graduated from the Mathematics Department of Sichuan University , Bachelor of Science . A series of research results were published in ACM SIGGRAPH\TOG、IEEE TPAMI、IEEE TVCG、NeurIPS and ICCV On . He has won a national scholarship , Time essence intelligence CAD&CG Excellent student award , The fourth paradigm doctoral award, etc .

Share content ：

For complex geometric structures and fine geometric details

Recursive depth generation model of

Introduction to the report ：

3D Geometry representation and modeling is one of the core tasks of computer graphics and computer vision , Understand from the scene 、 The algorithm design of object recognition and classification applied to shape reconstruction and editing depends on good 3D geometric representation and high-quality models . Ideally , The composite shape should be able to contain fine geometric details and complex structures , And the generation process needs to provide high-level control to ensure that the models we need are generated . The existing representation methods for deep geometry learning can not meet the above requirements at the same time , At the same time, controllable model synthesis is not possible . This work benefits from the advantages of hierarchical recursive graph representation and strong expression ability of homeomorphic deformation details . According to the internal multi-layer recursive structure design of the three-dimensional model, the corresponding depth generation model is proposed , A structured modeling method is introduced to represent and model the general three-dimensional artificial model . meanwhile , It can also decouple the geometric details and structure of the 3D model , And it can reasonably use the distribution of data for controllable model interpolation and generation . further , We use the structure of hierarchical recursive graph to express the layout and details of complex scenes , At the same time, the concept of functional area and Hyper edges Relationships to simplify the complex layout of the scene , Finally, a reasonable scene layout can be generated , Including the generation and reconstruction of fine component details of objects . Relevant work has been published in top journals of graphics ACM SIGGRAPH\TOG And the top issue of artificial intelligence IEEE TPAMI On .

Liufenglin ：

Institute of computing technology, Chinese Academy of Sciences 2021 Master's degree , The tutors are associate researcher lichunpeng and associate researcher Gaolin , His research interests are computer graphics , Relevant research work was published SIGGRAPH On .

Share content ：

Sketch based face image and video editing

Introduction to the report ：

Existing face image and video editing methods , Need to use complex software and professional skills , Higher editing costs . For the above problems , We have developed a sketch based image editing system DeepFaceEditing And video editing system DeepFaceVideoEditing, The two works were published in the graphics summit SIGGRAPH2021 and SIGGRAPH2022, And included in TOG Journal long article .DeepFaceEditing A structural decoupling framework for human face is proposed , Use a local to global blocking strategy , Decouple the geometry and appearance of the face . The system can use sketch to modify the geometric information of human face , Edit facial features, hair structure, etc , The appearance information of the face can also be modified based on the reference image , Change the skin color and hair color of the face . further , We propose the first sketch based face video editing system DeepFaceVideoEditing, Use StyleGAN Generation network , Project face video into hidden space . User's sketch editing operations are abstractly represented as editing vectors , Time independent and time dependent propagation . meanwhile , The system uses a regional fusion strategy , Support users in any frame / Multi frame input different editing operations .

Yuanyujie :

Doctoral candidate of Institute of computing technology, Chinese Academy of Sciences , The instructor is associate researcher Gaolin . His research interests include computer graphics and 3D computer vision , Bachelor degree from Xi'an Jiaotong University , Bachelor of Science . Research results published in ACM SIGGRAPH\TOG、CVPR On .

Share content ：

Geometry and appearance editing method of neural radiation field

Introduction to the report ：

In recent years , Neural rendering technology is developing rapidly . Neural radiation field is one of the most concerned works , Excellent performance in the new perspective synthesis task . However, the neural radiation field as an implicit representation , It is difficult to edit its geometry and appearance under user control . This report will focus on the editing method of neural radiation field , A geometric editing method and an appearance editing method of neural radiation field are introduced respectively . We first propose an interactive neural radiation field geometric deformation method , Establish a relationship between explicit triangular mesh representation and implicit representation , Deformation method using explicit triangular mesh , Realize the deformation editing of implicitly encoded geometry under the control of users , The edited results support free view browsing . Edit for appearance , We propose a neural radiation field based 3D Scene stylization method , Take advantage of 2D-3D Mutual learning methods , Let the neural radiation field network and 2D The convolutional decoding networks make use of each other's advantages , It achieves both consistency and high-quality stylization . Two studies were published in CVPR 2022 On .

6 month 9 Japan 19:30-20:30

Wu Tong :

Doctoral candidate of Institute of computing technology, Chinese Academy of Sciences , His research interests are computer graphics and 3D computer vision , The related research work is published in SIGGRAPH On .

Share content ：

A texture 3D model generation network TM-NET

Introduction to the report ：

In recent years , The generation network has achieved great success in the field of images and 3D models , However, the existing point cloud 、 Voxel 、 The effect of texture 3D model generated by implicit field and mesh is not ideal , Unable to express high-frequency texture details . This report will introduce a high-quality texture 3D model generation method based on structured bounding box expansion , Expand the bounding box as a unified texture mapping domain , Encode high-frequency details of texture by quantizing encoder , The relationship between geometry and texture is established by conditional autoregressive model , High quality texture 3D model generation is realized .

Sunyangtian :

Graduate student of Institute of computing technology, Chinese Academy of Sciences , The tutor is Gao Lin, associate researcher . His research interests are computer graphics and 3D computer vision , The related research work is published in IEEE TPAMI and CVPR On .

Share content ：

Human video rendering based on dynamic texture representation

Introduction to the report ：

The generation of digital human has received extensive attention in recent years . However , Methods based on image translation technology often need a large number of human videos for training , The generalization ability is limited ; The work based on rendering can not better express the dynamic details of human surface . This report introduces a synthesis method of dynamic high-frequency details during human body rendering , By embedding graphics translation technology into the micro rendering framework , It realizes the synthesis of human dynamic details with high realism .

After the live broadcast, you can ask questions in the group , Please add “AI TIME Little helper （ WeChat ID ：AITIME_HY）”, reply “ graphics ”, Will pull you into “AI TIME Computer vision communication group ”！

AI TIME Wechat assistant

Lord         do ：AI TIME 

Associated Media ：AI Data pie 、 Academic headlines

partners ： Wisdom spectrum ·AI、 Chinese Academy of Engineering Zhiling live 、 School Online 、 Kou enjoys academic 、AMiner、 Ever Chain action 、 Scientific research cloud

Excellent articles in the past are recommended

Remember to pay attention to us ！ There is new knowledge every day ！

  About AI TIME 

AI TIME From 2019 year , It aims to carry forward the spirit of scientific speculation , Invite people from all walks of life to the theory of artificial intelligence 、 Explore the essence of algorithm and scenario application , Strengthen the collision of ideas , Link the world AI scholars 、 Industry experts and enthusiasts , I hope in the form of debate , Explore the contradiction between artificial intelligence and human future , Explore the future of artificial intelligence .

so far ,AI TIME Has invited 600 Many speakers at home and abroad , Held more than 300 An event , super 170 10000 people watch .

I know you.

Looking at

Oh

~

Click on Read the original   Reservation live broadcast ！