Don't Worry About the Weather: Unsupervised Condition-Dependent Domain Adaptation
This addresses the issue of robustness in system-critical tasks like segmentation and localization for applications in autonomous driving or robotics, though it is incremental as it builds on existing domain adaptation methods.
The paper tackles the problem of performance degradation in computer vision models when input conditions change, by introducing a domain adaptation system that uses light-weight input adapters to pre-process images for compatibility with off-the-shelf models trained on ideal conditions, reporting large improvements in semantic segmentation and topological localization on RobotCar and BDD datasets.
Modern models that perform system-critical tasks such as segmentation and localization exhibit good performance and robustness under ideal conditions (i.e. daytime, overcast) but performance degrades quickly and often catastrophically when input conditions change. In this work, we present a domain adaptation system that uses light-weight input adapters to pre-processes input images, irrespective of their appearance, in a way that makes them compatible with off-the-shelf computer vision tasks that are trained only on inputs with ideal conditions. No fine-tuning is performed on the off-the-shelf models, and the system is capable of incrementally training new input adapters in a self-supervised fashion, using the computer vision tasks as supervisors, when the input domain differs significantly from previously seen domains. We report large improvements in semantic segmentation and topological localization performance on two popular datasets, RobotCar and BDD.