page contents

NetEase AI Lab wins double championship in the world's top AI conference IJCAI 3D Challenge

Recently, NetEase AI Lab stood out in the 3D AI Challenge of IJCAI 2020, the world's top artificial intelligence conference, and achieved a significant lead in second place. It won the double championship of 3D object reconstruction and image-based 3D object retrieval. Netease's leading technical strength in the field of 3D vision.



This competition is based on the industry’s first industrial large-scale 3D furniture data set (3D-FUTURE). It is divided into three tracks: object reconstruction, image-based 3D object retrieval, and instance segmentation. It aims to explore how algorithms are used in 3D object understanding and object generation. which performed.

Relying on the technology accumulated in the 3D field, NetEase AI Lab has carried out innovative designs in network structure design and data enhancement.

In the track of 3D object reconstruction, NetEase AI Lab adopts a non-uniform sparse sampling strategy of point clouds, and uses rendering technology to enhance the data from multiple perspectives. It is evaluated in Chamfer Distance and average F-score. The index significantly leads the second place to win the championship.

In the 3D object retrieval track, Netease AI Lab cleverly designed a new classification label based on the object similarity distance, and proposed a corresponding model fusion strategy. The final submitted model achieved a high score of 57.06, which is higher than the second place. The team is 11% higher.

While tackling technical difficulties, Netease AI Lab continues to explore the application of 3D vision technology in various fields. Video motion capture, facial expression capture, face fusion, and stylized reconstruction of head models have all been applied in actual scenes, which not only reduces the production cost of products, but also enriches the user experience.

3D vision is a basic research field that has received wide attention in the field of artificial intelligence, and it is the only way to build a future 3D intelligent world. In the top academic conferences in recent years, the number of papers on 3D vision topics has gradually increased and is at the forefront of research.

Source: NetEase Technology Report, translated by Google Translate


  • QQ空间

  • 腾讯微博

  • 新浪微博

  • 人人网

  • 豆瓣

技术支持: 机器人行业建站
  • Home
  • Call
  • Msg
  • add.
  • Vip