Kangle Deng

I am a PhD student at the Robotics Institute of Carnegie Mellon University, where I'm fortunate to be co-advised by Deva Ramanan and Jun-Yan Zhu.

Prior to joining CMU, I obtained my B.S degree in 2020 from Peking University.

So far, my research mainly involves computer-aided creation. My current work is supported by Microsoft Research PhD Fellowship.

Email / GitHub / Google Scholar

Selected Publications

	Tactile DreamFusion: Exploiting Tactile Sensing for 3D Generation Ruihan Gao, Kangle Deng, Gengshan Yang, Wenzhen Yuan, Jun-Yan Zhu NeurIPS, 2024 project page / github 3D content creation with touch: TactileDreamFusion exploits high-resolution tactile sensing to enhance geometric details for text- or image-to-3D generation.
	FlashTex: Fast Relightable Mesh Texturing with LightControlNet Kangle Deng, Timothy Omernick, Alexander Weiss, Deva Ramanan, Jun-Yan Zhu, Tinghui Zhou, Maneesh Agrawala ECCV, 2024 (Oral Presentation) project page / github FlashTex textures an input 3D mesh given a user-provided text prompt. Notably, our generated texture can be relit properly in different lighting environments.
	StableLego: Stability Analysis of Block Stacking Assembly Ruixuan Liu, Kangle Deng, Ziwei Wang, Changliu Liu IEEE Robotics and Automation Letters (RA-L), 2024 github We propose a new optimization formulation to infer the structural stability of block stacking assembly. We also provide StableLego: a comprehensive Lego assembly dataset of more than 50k Lego structures with their stability inferences.
	Total-Recon: Deformable Scene Reconstruction for Embodied View Synthesis Chonghyuk(Andrew) Song, Gengshan Yang, Kangle Deng, Jun-Yan Zhu, Deva Ramanan ICCV, 2023 project page / github Given a RGBD video of deformable objects, Total-Recon builds 3D models of objects and background, which enables embodied view synthesis.
	3D-aware Conditional Image Synthesis Kangle Deng, Gengshan Yang, Deva Ramanan, Jun-Yan Zhu CVPR, 2023 project page / github We propose pix2pix3D, a 3D-aware conditional generative model for controllable photorealistic image synthesis. Given a 2D label map, such as a segmentation or edge map, our model learns to synthesize a corresponding image from different viewpoints.
	Depth-supervised NeRF: Fewer Views and Faster Training for Free Kangle Deng, Andrew Liu, Jun-Yan Zhu, Deva Ramanan CVPR, 2022 project page / github Proposed DS-NeRF (Depth-supervised Neural Radiance Fields), a model for learning neural radiance fields that takes advantage of depth supervised by 3D point clouds.
	Unsupervised Any-to-Many Audiovisual Synthesis via Exemplar Autoencoders Kangle Deng, Aayush Bansal, Deva Ramanan ICLR, 2021 project page Defined and addressed a new question of unsupervised audiovisual synthesis -- input the audio of a random individual and then output the talking-head video with audio in the style of another target speaker.
	IRC-GAN: Introspective Recurrent Convolutional GAN for Text-to-video Generation Kangle Deng, Tianyi Fei, Xin Huang, Yuxin Peng IJCAI, 2019 bibtex Application of Mutual Information to Text-to-video Generation.