HTTP/2 200 server: GitHub.com content-type: text/html; charset=utf-8 last-modified: Tue, 16 Sep 2025 12:29:07 GMT access-control-allow-origin: * strict-transport-security: max-age=31556952 etag: W/"68c95813-4ade" expires: Mon, 29 Dec 2025 19:48:01 GMT cache-control: max-age=600 content-encoding: gzip x-proxy-cache: MISS x-github-request-id: CD76:3ABDEF:93030F:A506FE:6952D897 accept-ranges: bytes age: 0 date: Mon, 29 Dec 2025 19:38:01 GMT via: 1.1 varnish x-served-by: cache-bom-vanm7210058-BOM x-cache: MISS x-cache-hits: 0 x-timer: S1767037082.562140,VS0,VE220 vary: Accept-Encoding x-fastly-request-id: bee60f3231d8bc7abe23f421e6528a739588a916 content-length: 4310

Surfel-based Gaussian Inverse Rendering for Fast and Relightable Dynamic Human Reconstruction from Monocular Videos

Yiqun Zhao¹, Chenming Wu², Binbin Huang¹, Yihao Zhi³, Chen Zhao², Jingdong Wang², Shenghua Gao^1†,

¹The University of Hong Kong ²Baidu Inc. ³The Chinese University of Hong Kong, Shenzhen

(† denotes the corresponding author)

arXiv

SGIA: We achieve fast reconstruction of clothed human avatars with PBR properties from a monocular video. SGIA takes a monocular video and initial human pose and shape as input to estimate dynamic clothed humans' PBR properties, including geometry and materials. Leveraging a PBR-aware 2DGS representation, our method enables fast training and rendering processes. By utilizing the estimated PBR properties, we can not only deform the avatars into different poses but also render them with realistic lighting conditions, allowing for versatile and visually appealing outputs.

Main Experiment Comparison

Select Training/Novel pose to compare.

Training pose Novel Pose

(Please refresh if you don't see the video. Switch video will need some time to process.)

Baselines    VS    Ours

Novel Pose Animation at different View under different illumination

(Please refresh if you don't see the video. Switch video will need some time to process.)

Male-3 Male-4 Female-3 Female-4

(Click here to switch the Subject)

Novel lighing results on zju mocap dataset