Tobias Fischer's Avatar

Tobias Fischer

@tobiasfshr.bsky.social

Research Scientist Intern @ NVIDIA PhD Student @ CVG, ETH Zurich Prev. @ Meta Reality Labs, UC Berkeley, RWTH Dynamic 3D Vision, NeRFs, 3DGS tobiasfshr.github.io

113 Followers  |  175 Following  |  5 Posts  |  Joined: 27.11.2024  |  1.5967

Latest posts by tobiasfshr.bsky.social on Bluesky

Post image

The village of Blatten in Switzerland is partly buried by the collapse of a glacier which melted because our greenhouse gas emissions warmed the climate.

Another reminder that choosing to not invest enough today to reduce emissions means far greater costs for society in the long term.

29.05.2025 09:16 β€” πŸ‘ 346    πŸ” 139    πŸ’¬ 8    πŸ“Œ 13
Post image

If youβ€˜re watching #eurovision tonight, look out for the robots from ETH!
Really cool to see something I could work with during my PhD featured as a swiss highlight πŸ€–

17.05.2025 19:37 β€” πŸ‘ 9    πŸ” 1    πŸ’¬ 0    πŸ“Œ 0
Preview
FlowR: Flowing from Sparse to Dense 3D Reconstructions | Tobias Fischer We present FlowR, a novel pipeline that bridges the gap between sparse and dense 3D reconstruction. Contrary to prior works, we learn a direct mapping between incorrect renderings and their correspond...

Project Page: tobiasfshr.github.io/pub/flowr/

03.04.2025 14:48 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0
Video thumbnail

3/3 While (a) 'FlowR (Initial)' helps to disambiguate geometry, many artifacts remain. FlowR fixes these with a multi-view, flow matching model that maps incorrect renderings to ground truth (b).

03.04.2025 09:34 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

2/3 FlowR consists of two parts:
(a) A robust initial reconstruction pipeline based on MASt3R/3DGS, designed for both sparse- and dense-view settings.
(b) A data densification procedure using flow matching to generate high-quality extra views that improve reconstruction.

03.04.2025 09:33 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0
Video thumbnail

1/3 Introducing FlowR 🌸: Flowing from Sparse to Dense 3D Reconstructions

We learn a direct mapping between incorrect renderings and their corresponding ground-truth images, augmenting scene captures with consistent novel, generated views to improve reconstruction quality.

03.04.2025 09:33 β€” πŸ‘ 2    πŸ” 1    πŸ’¬ 1    πŸ“Œ 0

I wonder what is the position of the CVF w.r.t. the possibility of such events regarding upcoming conferences in the USA.

19.03.2025 19:46 β€” πŸ‘ 17    πŸ” 3    πŸ’¬ 0    πŸ“Œ 0
Post image

Thatβ€˜s a wrap! Thanks to the organizers of @neuripsconf.bsky.social for putting together an amazing conference. See you soon Vancouver (what a beautiful venue!)

15.12.2024 23:15 β€” πŸ‘ 4    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

@tobiasfshr is following 20 prominent accounts