Enlighten Games With AI

NeuroSkinning: Two Datasets for Automatic Skin Binding


  We introduce two large skinning datasets for research. These two datasets are collected from our two popular online games, the examples are created and rigged with our professional artists. In both of the two datasets, an example consists of a character model, the corresponding skeleton structure and the skinning weights. The Game A dataset consists of about 500 characters, all the models wear complex, ancient costumes. The number of vertices of a character ranges from 5,000 to 40,000, and most mehses are non-manifold and composed of multiple disjoint components. All characters are in rest pose and corresponding to the same skeleton structure and a vertex is set to bu influenced by no more than 4 bones. The Game B dataset has 24 different skeleton structures, each of which is shared by 20-50 characters, resulting in 1171 characters in total. The number of a character model ranges from 1,000 to 5,000, and the number of bones ranges from 80 to 230. Some examples are showed below. The two datasets can be used for data-driven skin binding research as shown in the reference paper, and also have the potential to benefit a wide range of other applications.

Figure 1. 10 character examples in the game ‘A’ dataset.

Figure 2. 10 character examples in the game ‘B’ dataset. Top level: the character models. Bottom level: the corresponding skeleton structures.

Obtaining the data

  We make the two datasets available for academic research purpose. To obtain a copy, please send an email to stating:

  (1) your name, title, affiliation (if you are a student, please ask your advisor to contact us)

  (2) your intended use of the data

  (3) a statement saying that you accept the following terms of licensing:

    1. Non-commercial Use: The license granted is for internal, non-commercial research, evaluation or testing purposes only. Any use of the datasets or its contents to manufacture or sell products or technologies (or portions thereof) either directly or indirectly for any direct or indirect for-profit purposes is strictly prohibited.

    2. Controlled Distribution: The rights to copy, distribute, and use the data you are being given access to are under the control of Netease. You are hereby given permission to copy this data in electronic or hardcopy form for your own scientific use and to distribute it for scientific use to colleagues with your research group. Inclusion of rendered images or video made from this data in a scholarly publication (printed or electronic) is also permitted. However, the data may not be included in the electronic version of a publication, nor placed on the Internet. These restrictions apply to any representations (other than images or video) derived from the data, including but not limited to simplifications, remeshing, and the fitting of smooth surfaces. The making of physical replicas this data is prohibited, and the data may not be distributed to students in connection with a class. For any other use, including distribution outside your research group, written permission is required from Netease.

    3. Citation: You agree to referenece to the publicaton "NeuroSkinning: Automatic Skin Binding for Production Characters with Deep Graph Networks" and acknowledge Netease as the source of the Datasets in any publications reporting use of it or any manual or document. A copy of all reports and papers that are for public or general release that use the datasets must be forwarded immediately upon release or publication to Netease.

File formats

  All the examples of the two datasets are stored in FBX format. Please notice that all the examples in the Game A dataset are in rest pose, and some examples in the Game B dataset are in non-rest pose.


  Lijuan Liu, Youyi Zheng, Di Tang, Yi Yuan, Changjie Fan, and Kun Zhou. 2019. NeuroSkinning: Automatic Skin Binding for Production Characters with Deep Graph Networks. ACM Trans. Graph. 38, 4, Article 1 (July 2019), 13 pages. 3322969