This paper studies the task of image captioning with novel objects, which only exist in testing images. Intrinsically, this task can reflect the generalization ability of models in understanding and captioning the semantic meanings of visual concepts and objects unseen in training set, sharing the similarity to one/zero-shot learning. The critical difficulty thus comes from that no paired images and sentences of the novel objects can be used to help train the captioning model. Inspired by recent work (Chen et al. 2019b) that boosts one-shot learning by learning to generate various image deformations, we propose learning meta-networks for deforming features for novel object captioning. To this end, we introduce the feature deformation meta-networks (FDM-net), which is trained on source data, and learn to adapt to the novel object features detected by the auxiliary detection model. FDM-net includes two sub-nets: feature deformation, and scene graph sentence reconstruction, which produce the augmented image features and corresponding sentences, respectively. Thus, rather than directly deforming images, FDM-net can efficiently and dynamically enlarge the paired images and texts by learning to deform image features. Extensive experiments are conducted on the widely used novel object captioning dataset, and the results show the effectiveness of our FDM-net. Ablation study and qualitative visualization further give insights of our model.

This paper presents a framework for real-time high-fidelity upper limb motion transfer from a human to a humanoid robot, intending to provide robust support for robots to learn manipulation skills. Robot imitation of human motion is modeled as a simultaneous wrist and elbow tracking problem, where the coordination between multiple imitation objectives and physical constraints receives special attention. First, for the vision-based motion capture system, an efficient preprocessing method is introduced to map the raw human skeletal points analogously or reflectively to the robot’s workspace. Then, the generation of velocity-level imitation motion commands is formalized as a constrained hierarchical quadratic programing (HQP) problem, in which multiple objectives for achieving imitation are organized into a hierarchy. The wrist tracking task is set as the highest priority, followed by the adjustment of the elbow and the minimization of the joint velocity. Physical limits of the robot joints, as well as strict kinematic constraints regarding self-motion, are considered as constraints for the optimization problem. Finally, a hierarchical recurrent neural network (RNN) is designed to solve this HQP problem through its unique neural dynamics. The performance of this network is theoretically guaranteed with high computational efficiency. Quantitative results show that the proposed motion imitation control scheme offers excellent integrated tracking accuracy of [ ] m and is robust to different arm length ratios. Using this method, the humanoid robot can imitate the upper body motions of a human as accurately as possible under physical constraints, without significant delays.