Predicting the grasping point accurately and quickly is crucial for successful robotic manipulation. However, to commercially deploy a robot, such as a dishwasher robot in a commercial kitchen, we also need to consider the constraints of limited usable resources. We present a deep learning method to predict the grasp position when using a single suction gripper for picking up objects. The proposed method is based on a shallow network to enable lower training costs and efficient inference on limited resources. Costs are further reduced by collecting data in a custom-built synthetic environment. For evaluating the proposed method, we developed a system that models a commercial kitchen for a dishwasher robot to manipulate symmetric objects. We tested our method against a model-fitting method and an algorithm-based method in our developed commercial kitchen environment and found that a shallow network trained with only the synthetic data achieves high accuracy. We also demonstrate the practicality of using a shallow network in sequence with an object detector for ease of training, prediction speed, low computation cost, and easier debugging.

In this paper we propose an extension of the Attention Branch Network (ABN) by using instance segmentation for generating sharper attention maps for action recognition. Methods for visual explanation such as Grad-CAM usually generate blurry maps which are not intuitive for humans to understand, particularly in recognizing actions of people in videos. Our proposed method, Object-ABN, tackles this issue by introducing a new mask loss that makes the generated attention maps close to the instance segmentation result. Further the Prototype Conformity (PC) loss and multiple attention maps are introduced to enhance the sharpness of the maps and improve the performance of classification. Experimental results with UCF101 and SSv2 shows that the generated maps by the proposed method are much clearer qualitatively and quantitatively than those of the original ABN.

Action spotting is a key component in high-level video understanding. The large number of similar frames poses a challenge for recognizing actions in videos. In this paper we use frame saliency to represent the importance of frames for guiding the model to focus on keyframes. We propose the frame saliency weighting module to improve frame saliency and video representation at the same time. Our proposed model contains two encoders, for pre-action and post-action time windows, to encode video context. We validate our design choices and the generality of proposed method in extensive experiments. On the public SoccerNet-v2 dataset, the method achieves an average mAP of 57.3%, improving over the state of the art. Using embedding features obtained from multiple feature extractors, the average mAP further increases to 75%. We show that reducing the model size by over 90% does not significantly impact performance. Additionally, we use ablation studies to prove the effective of saliency weighting module. Further, we show that our frame saliency weighting strategy is applicable to existing methods on more general action datasets, such as SoccerNet-v1, ActivityNet v1.3, and UCF101.