Anton Obukhov

toshas

AI & ML interests

None yet

Organizations

Posts 4

view post
Post
951
Join us at our remaining CVPR presentations this week! Members of PRS-ETH will be around to connect with you and discuss our presented and ongoing works:

๐Ÿ’ Marigold: Discover our work on sharp diffusion-based computer vision techniques, presented in Orals 3A track on "3D from Single View", Thu, June 20, 9:00-9:15 AM. Also, drop by Poster Session 3 later that day for more tangible matters! ๐ŸŒš
Project page: https://marigoldmonodepth.github.io/
Paper: Repurposing Diffusion-Based Image Generators for Monocular Depth Estimation (2312.02145)
Collection: https://huggingface.co/collections/prs-eth/marigold-6669e9e3d3ee30f48214b9ba
Space: prs-eth/marigold-lcm
Diffusers ๐Ÿงจ tutorial: https://huggingface.co/docs/diffusers/using-diffusers/marigold_usage

โš™๏ธ Point2CAD: Learn about our mechanical CAD model reconstruction from point clouds, presented in Poster Session 1, Wed, June 19, 10:30 AM - 12:00 PM.
Project page: https://www.obukhov.ai/point2cad.html
Paper: Point2CAD: Reverse Engineering CAD Models from 3D Point Clouds (2312.04962)

๐ŸŽญ DGInStyle: Explore our generative data synthesis approach as a cost-efficient alternative to real and synthetic data, presented in the Workshop on Synthetic Data for Computer Vision, Tue, June 18, at Summit 423-425.
Details and schedule: https://syndata4cv.github.io/
Project page: https://dginstyle.github.io/
Paper: DGInStyle: Domain-Generalizable Semantic Segmentation with Image Diffusion Models and Stylized Semantic Control (2312.03048)
Model: yurujaja/DGInStyle
view post
Post
1951
Another gem from our lab โ€” DGInStyle! We use Stable Diffusion to generate semantic segmentation data for autonomous driving and train domain-generalizable networks.

๐Ÿ“Ÿ Website: https://dginstyle.github.io
๐Ÿงพ Paper: https://arxiv.org/abs/2312.03048
๐Ÿค— Hugging Face Paper: DGInStyle: Domain-Generalizable Semantic Segmentation with Image Diffusion Models and Stylized Semantic Control (2312.03048)
๐Ÿค— Hugging Face Model: yurujaja/DGInStyle
๐Ÿ™ Code: https://github.com/yurujaja/DGInStyle

In a nutshell, our pipeline overcomes the resolution loss of Stable Diffusion latent space and the style bias of ControlNet, as shown in the attached figures. This allows us to generate sufficiently high-quality pairs of images and semantic masks to train domain-generalizable semantic segmentation networks.

Team: Yuru Jia ( @yurujaja ), Lukas Hoyer, Shengyu Huang, Tianfu Wang ( @Tianfwang ), Luc Van Gool, Konrad Schindler, and Anton Obukhov ( @toshas ).

models

None public yet

datasets

None public yet