Papers of the day   All papers

Towards Robust Monocular Depth Estimation: Mixing Datasets for Zero-shot Cross-dataset Transfer

Comments

Ankur Handa: Impressive results on single image depth estimation by Ranftl et al https://arxiv.org/abs/1907.01341 https://t.co/f6nrhXyCCH

4 replies, 414 likes


hardmaru: The coolest result in this paper is when they took a depth estimation model (single-image input) trained on natural images (https://arxiv.org/abs/1907.01341), and showed that the pre-trained model also works on certain types of line drawings, such as drawings of streets and indoor scenes. https://t.co/l7zywvDolj

8 replies, 252 likes


Daisuke Okanohara: We can combine multiple training datasets for monocular depth estimation with a scale-shift invariant loss and multi-objective loss. Also, 3D movies can be used as a training dataset. Achieved SOTA for monocular depth estimation. http://arxiv.org/abs/1907.01341 https://youtu.be/D46FzVyL9I8

0 replies, 5 likes


Sergey Matveev: Now is the time to make autonomous robots, because they can see now. Next step is understanding the context.

0 replies, 2 likes


HotComputerScience: Most popular computer science paper of the day: "Towards Robust Monocular Depth Estimation: Mixing Datasets for Zero-shot Cross-dataset Transfer" https://hotcomputerscience.com/paper/towards-robust-monocular-depth-estimation-mixing-datasets-for-zero-shot-cross-dataset-transfer https://twitter.com/ankurhandos/status/1220755115256057856

0 replies, 2 likes


akira: https://arxiv.org/abs/1907.01341 Because each dataset of depth estimation has different camera parameters, a model trained with one dataset may not be accurate in another dataset. They propose a dataset-independent loss function to prevent accuracy degradation when crossing datasets. https://t.co/Wilb6wTibB

0 replies, 2 likes


akira: https://arxiv.org/abs/1907.01341 Since each dataset for depth has different accuracy and camera parameters, the accuracy of the model learned with one model drops at another dataset. The authors proposed a dataset-independent loss function and used it prevents accuracy degradation. https://t.co/SiL54OPp0B

0 replies, 1 likes


arXiv CS-CV: Towards Robust Monocular Depth Estimation: Mixing Datasets for Zero-Shot Cross-Dataset Transfer http://arxiv.org/abs/1907.01341

0 replies, 1 likes


Content

Found on Jan 24 2020 at https://arxiv.org/pdf/1907.01341.pdf

PDF content of a computer science paper: Towards Robust Monocular Depth Estimation: Mixing Datasets for Zero-shot Cross-dataset Transfer