Towards Robust Monocular Depth Estimation: Mixing Datasets for Zero-shot Cross-dataset Transfer
Paper
•
1907.01341
•
Published
DPT 3.1 release
DPT 3.1 (MiDaS) models, leveraging state-of-the-art vision backbones such as BEiT and Swinv2
-
Intel/dpt-beit-large-512
Depth Estimation • Updated • 25.3k • 7Note This model gives the highest quality, but is also the most heavy in terms of computation as mentioned in the paper.
-
Intel/dpt-beit-large-384
Depth Estimation • Updated • 84 -
Intel/dpt-beit-base-384
Depth Estimation • Updated • 56.2k • 1
Intel/dpt-swinv2-large-384
Depth Estimation • Updated • 106Note This model has moderately less quality, but has a better speed-performance trade-off
-
Intel/dpt-swinv2-base-384
Depth Estimation • Updated • 43
Intel/dpt-swinv2-tiny-256
Depth Estimation • Updated • 578 • 7Note This model is recommended for deployment on embedded devices