441 papers with code • 10 benchmarks • 55 datasets
Depth Estimation is the task of measuring the distance of each pixel relative to the camera. Depth is extracted from either monocular (single) or stereo (multiple views of a scene) images. Traditional methods use multi-view geometry to find the relationship between the images. Newer methods can directly estimate depth by minimizing the regression loss, or by learning to generate a novel view from a sequence. The most popular benchmarks are KITTI and NYUv2. Models are typically evaluated according to a RMS metric.
Accurate depth estimation from images is a fundamental task in many applications including scene understanding and reconstruction.
On the other hand, epistemic uncertainty accounts for uncertainty in the model -- uncertainty which can be explained away given enough data.
In particular, we propose a robust training objective that is invariant to changes in depth range and scale, advocate the use of principled multi-objective learning to combine data from different sources, and highlight the importance of pretraining encoders on auxiliary tasks.
We show that the proposed method outperforms the state-of-the-art works with significant margin evaluating on challenging benchmarks.