Timezone: »

 
Poster
Dynamo-Depth: Fixing Unsupervised Depth Estimation for Dynamical Scenes
Yihong Sun · Bharath Hariharan

Wed Dec 13 03:00 PM -- 05:00 PM (PST) @ Great Hall & Hall B1+B2 #215
in Poster Session 4 »
Event URL: https://dynamo-depth.github.io »

Unsupervised monocular depth estimation techniques have demonstrated encouraging results but typically assume that the scene is static. These techniques suffer when trained on dynamical scenes, where apparent object motion can equally be explained by hypothesizing the object's independent motion, or by altering its depth. This ambiguity causes depth estimators to predict erroneous depth for moving objects. To resolve this issue, we introduce Dynamo-Depth, an unifying approach that disambiguates dynamical motion by jointly learning monocular depth, 3D independent flow field, and motion segmentation from unlabeled monocular videos. Specifically, we offer our key insight that a good initial estimation of motion segmentation is sufficient for jointly learning depth and independent motion despite the fundamental underlying ambiguity. Our proposed method achieves state-of-the-art performance on monocular depth estimation on Waymo Open and nuScenes Dataset with significant improvement in the depth of moving objects. Code and additional results are available at https://dynamo-depth.github.io.

Author Information

Yihong Sun (Cornell University)
Yihong Sun

I am a CS PhD Student at Cornell University, advised by Prof. Bharath Hariharan. Previously, I obtained my Bachelor’s degree from Johns Hopkins University where I worked with Bloomberg Distinguished Prof. Alan Yuille and Dr. Adam Kortylewski. My work is supported by the NSF GRFP and my research interests are computer vision and machine learning, especially in building vision algorithms that can learn from little supervision and generalize to unseen domains.

Bharath Hariharan (Cornell University)

More from the Same Authors