The authors developed FOCAL, a geometric reward that quantifies cross-view depth consistency in static scene regions using depth maps from a pretrained 4D reconstructor.
Calibrated depth reward improves 4D consistency without harming visual quality
GeoFix applies a field-of-view-calibrated geometric reward to both training and inference-time alignment, achieving gains in geometric consistency and video quality simultaneously.
Big Tech
Yeobin Hong · Hyojun Go · Dohun Lee · Jeongjae Lee · Sunghwan Hong · Federico Tombari · +4 more
KAIST · ETH Zürich · Google · EPFL
Research Digest··2 min read
The authors propose GeoFix, a framework that uses a single geometric reward (FOCAL) to improve geometric consistency and visual quality in 4D generation from video diffusion models.
Why this paper
From Google and 3 others
In one line
A single calibrated geometric reward improves both 4D consistency and video quality without auxiliary rewards.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks (2 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§