Published in Submission, 2024
We reframe scene flow as the task of estimating a continuous space-time ODE that describes motion for an entire observation sequence, represented with a neural prior. Our method, EulerFlow, optimizes this neural prior estimate against several multi-observation reconstruction objectives, enabling high quality scene flow estimation via pure self-supervision on real-world data.
Download here
Published in European Conference on Computer Vision (ECCV), 2024
Current scene flow methods broadly fail to describe motion on small objects, and current scene flow evaluation protocols hide this failure by averaging over many points, with most drawn larger objects. To fix this evaluation failure we propose a new evaluation protocol: Bucket Normalized EPE, and a frustratingly simple supervised scene flow baseline to achieve SOTA: TrackFlow.
Download here
Published in International Conference on Learning Representations (ICLR), 2024
We propose Scene Flow via Distillation, a simple distillation framework that uses a label-free optimization method to produce pseudo-labels to supervise a feed forward model. Our technique yields state of the art results for endpoint error while still running 100x faster than the previously best performing methods.
Download here
Published in arXiv pre-print, 2023
We propose a method to perform end-to-end learning on raw radar analog-to-digital (ADC) data from imaging radars. Specifically, we design a learnable signal processing module inside the neural network, and a pre-training method guided by traditional signal processing algorithms. The combination of these techniques allows us to achieve state of the art results on the RADIal raw radar dataset.
Download here
Published in International Confderence on Intelligent Robots and Systems (IROS), 2021
A framework for calculating indirect estimations of state given observations of the behavior of other agents in the environment. We model traffic light estimation as such a problem and apply our framework to recover light state with over 70% accuracy using only the motion of other vehicles in the scene.
Download here