Domain Adaptation Through Task Distillation
arXiv:2008.11911 · doi:10.1007/978-3-030-58574-7_40
Abstract
Deep networks devour millions of precisely annotated images to build their complex and powerful representations. Unfortunately, tasks like autonomous driving have virtually no real-world training data. Repeatedly crashing a car into a tree is simply too expensive. The commonly prescribed solution is simple: learn a representation in simulation and transfer it to the real world. However, this transfer is challenging since simulated and real-world visual experiences vary dramatically. Our core observation is that for certain tasks, such as image recognition, datasets are plentiful. They exist in any interesting domain, simulated or real, and are easy to label and extend. We use these recognition datasets to link up a source and target domain to transfer models between them in a task distillation framework. Our method can successfully transfer navigation policies between drastically different simulators: ViZDoom, SuperTuxKart, and CARLA. Furthermore, it shows promising results on standard domain adaptation benchmarks.
arXiv commentPublished in European Conference on Computer Vision (ECCV 2020)
References in corpus (11)
- ADVENT: Adversarial Entropy Minimization for Domain Adaptation in Semantic Segmentation
- Sim-to-Real Transfer of Robotic Control with Dynamics Randomization
- CyCADA: Cycle-Consistent Adversarial Domain Adaptation
- The Open Images Dataset V4: Unified image classification, object detection, and visual relationship detection at scale
- Revisiting Batch Normalization For Practical Domain Adaptation
- Driving Policy Transfer via Modularity and Abstraction
- Sim-to-Real Robot Learning from Pixels with Progressive Nets
- ChauffeurNet: Learning to Drive by Imitating the Best and Synthesizing the Worst
- Does computer vision matter for action?
- Sim2real transfer learning for 3D human pose estimation: motion to the rescue
- Identifying Unknown Instances for Autonomous Driving