Convolutional neural networks show remarkable results in classification but struggle with learning new things on the fly. We present a novel rehearsal-free approach, where a deep neural network is continually learning new unseen object categories without saving any data of prior sequences. Our approach is called RECALL, as the network recalls categories by calculating logits for old categories before training new ones. These are then used during training to avoid changing the old categories. For each new sequence, a new head is added to accommodate the new categories. To mitigate forgetting, we present a regularization strategy where we replace the classification with a regression. Moreover, for the known categories, we propose a Mahalanobis loss that includes the variances to account for the changing densities between known and unknown categories. Finally, we present a novel dataset for continual learning (HOWS-CL-25), especially suited for object recognition on a mobile robot, including 150,795 synthetic images of 25 household object categories. Our approach RECALL outperforms the current state of the art on CORe50 and iCIFAR-100 and reaches the best performance on HOWS-CL-25.
RECALL is a new approach for lifelong robot learning or continual learning. This work's paper was published at the IEEE International Conference on Intelligent Robots and Systems (IROS2022) in Japan. You can find more information and code in our repository: https://github.com/DLR-RM/RECALL And you can download our dataset at zenodo: https://zenodo.org/record/7054171
RECALL: Rehearsal-free Continual Learning for Object Classification (IROS 2022)
Your consent to the storage of data ('cookies') is required for the playback of this video on Youtube.com. You can view and change your current data storage settings at any time under privacy.
RECALL: Rehearsal-free Continual Learning for Object Classification (IROS 2022)
RECALL is a new approach for lifelong robot learning or continual learning. This work's paper was published at the IEEE International Conference on Intelligent Robots and Systems (IROS2022) in Japan. You can find more information and code in our repository: https://github.com/DLR-RM/RECALL And you can download our dataset at zenodo: https://zenodo.org/record/7054171
HOWS-CL-25 (Household Objects Within Simulation dataset for Continual Learning) is a synthetic dataset especially designed for object classification on mobile robots operating in a changing environment (like a household), where it is important to learn new, never seen objects on the fly. This dataset can also be used for other learning use-cases, like instance segmentation or depth estimation. Or where household objects or continual learning are of interest. HOWS contains 150795 RGB images containing 25 categories, over 925 instances of household objects, and corresponding normal, depth, and segmentation images. The dataset was created using Blenderproc.
