We propose a new problem formulation and a corresponding evaluation framework to advance research on unsupervised domain adaptation for semantic image segmentation. The overall goal is fostering the development of adaptive learning systems that will continuously learn, without supervision, in ever-changing environments. Typical protocols that study adaptation algorithms for segmentation models are limited to few domains, adaptation happens offline, and human intervention is generally required, at least to annotate data for hyper-parameter tuning. We argue that such constraints are incompatible with algorithms that can continuously adapt to different real-world situations. To address this, we propose a protocol where models need to learn online, from sequences of temporally correlated images, requiring continuous, frame-by-frame adaptation. We accompany this new protocol with a variety of baselines to tackle the proposed formulation, as well as an extensive analysis of their behaviors, which can serve as a starting point for future research.
On the road to online adaptation for semantic image segmentation
| Riccardo Volpi, Pau De Jorge, Gabriela Csurka, Diane Larlus |
| Conference on Computer Vision and Pattern Recognition (CVPR), New Orleans, Louisiana, USA, 21-24 June, 2022 |
Summary
Our core contribution is introducing the OASIS benchmark, to advance research in Online Adaptation for Semantic Image Segmentation.
“Standard” unsupervised domain adaptation is typically carried out offline, takes into account a limited number of domains and often required human annotation for model validation. These conditions are not well suited for a model released into the wild, which needs to adapt continuously, on the fly and without human intervention. For this reason, we study here the more challenging problem on continual and unsupervised online adaptation and propose an ad hoc benchmark (OASIS). Our benchmark is designed with three well-defined stages: pre-train, validation and deployment: in the first step, the model is pre-trained offline; in the validation phase, the online adaptation method of interest is validated on simulated sequences of temporally correlated frames; finally, in the deployment step the validated approach is released on real sequences recorded into the wild. In the final step, hyper-parameter tuning is no longer allowed. We believe that such scenario better reflects the needs of models released into the wild.
We implement a baseline suite with a broad set of methods at the intersection between unsupervised domain adaptation and continual learning. Furthermore, we propose techniques to overcome catastrophic forgetting. We carry thorough experimental analyses, putting the basis the future research on this topic.
This video shows how the NAVER LABS Europe continual learning online adaptation method (right), applied to image segmentation (classes listed in boxes), compares to image segmentation without adaptation (left) or with a Naive continual learning method (middle) which suffers from catastrophic forgetting (classes forgotten over time).
Demo
This work was part of the Lifelong representation learning chair of the MIAI research institute.
@InProceedings{volpi2022cvpr, author = {Volpi, Riccardo and de Jorge, Pau and Larlus, Diane and Csurka, Gabriela}, title = {On the Road to Online Adaptation for Semantic Image Segmentation}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)}, year = {2022} }
To make robots autonomous in real-world everyday spaces, they should be able to learn from their interactions within these spaces, how to best execute tasks specified by non-expert users in a safe and reliable way. To do so requires sequential decision-making skills that combine machine learning, adaptive planning and control in uncertain environments as well as solving hard combinatorial optimisation problems. Our research combines expertise in reinforcement learning, computer vision, robotic control, sim2real transfer, large multimodal foundation models and neural combinatorial optimisation to build AI-based architectures and algorithms to improve robot autonomy and robustness when completing everyday complex tasks in constantly changing environments.
The research we conduct on expressive visual representations is applicable to visual search, object detection, image classification and the automatic extraction of 3D human poses and shapes that can be used for human behavior understanding and prediction, human-robot interaction or even avatar animation. We also extract 3D information from images that can be used for intelligent robot navigation, augmented reality and the 3D reconstruction of objects, buildings or even entire cities.
Our work covers the spectrum from unsupervised to supervised approaches, and from very deep architectures to very compact ones. We’re excited about the promise of big data to bring big performance gains to our algorithms but also passionate about the challenge of working in data-scarce and low-power scenarios.
Furthermore, we believe that a modern computer vision system needs to be able to continuously adapt itself to its environment and to improve itself via lifelong learning. Our driving goal is to use our research to deliver embodied intelligence to our users in robotics, autonomous driving, via phone cameras and any other visual means to reach people wherever they may be.
This web site uses cookies for the site search, to display videos and for aggregate site analytics.
Learn more about these cookies in our privacy notice.
Cookie settings
You may choose which kind of cookies you allow when visiting this website. Click on "Save cookie settings" to apply your choice.
This content is currently blocked. To view the content please either 'Accept social media cookies' or 'Accept all cookies'.
For more information on cookies see our privacy notice.