Diverse Complexity Measures for Dataset Curation in Self-driving. (arXiv:2101.06554v1 [cs.LG])

Diverse Complexity Measures for Dataset Curation in Self-driving. (arXiv:2101.06554v1 [cs.LG])
https://ift.tt/2KsqWtD

Modern self-driving autonomy systems heavily rely on deep learning. As a consequence, their performance is influenced significantly by the quality and richness of the training data. Data collecting platforms can generate many hours of raw data in a daily basis, however, it is not feasible to label everything. It is thus of key importance to have a mechanism to identify "what to label". Active learning approaches identify examples to label, but their interestingness is tied to a fixed model performing a particular task. These assumptions are not valid in self-driving, where we have to solve a diverse set of tasks (i.e., perception, and motion forecasting) and our models evolve over time frequently. In this paper we introduce a novel approach and propose a new data selection method that exploits a diverse set of criteria that quantize interestingness of traffic scenes. Our experiments on a wide range of tasks and models show that the proposed curation pipeline is able to select datasets that lead to better generalization and higher performance.

via cs.RO updates on arXiv.org http://arxiv.org/

CoffeeKumazaki / arXiv

Diverse Complexity Measures for Dataset Curation in Self-driving. (arXiv:2101.06554v1 [cs.LG]) #2894