We propose encoder-centric stepwise models for extractive summarization using structured transformers - HiBERT (Zhang et al., 2019) and Extended Transformers (Ainslie et al., 2020). We enable stepwise summarization by injecting the previously generated summary into the structured transformer as an auxiliary sub-structure. Our models are not only efficient in modeling the structure of long inputs, but they also do not rely on task-specific redundancy-aware modeling, making them a general purpose extractive content planner for different tasks. When evaluated on CNN/DailyMail extractive summarization, stepwise models achieve state-of-the-art performance in terms of Rouge without any redundancy aware modeling or sentence filtering. This also holds true for Rotowire tableto-text generation, where our models surpass previously reported metrics for content selection, planning and ordering, highlighting the strength of stepwise modeling. Amongst the two structured transformers we test, stepwise Extended Transformers provides the best performance across both datasets and sets a new standard for these challenges. 1

This paper focuses on real-time rotation estimation for model-based automated visual inspection. In the case of model-based inspection, spatial alignment is essential to distinguish visual defects from normal appearance variations. Defects are detected by comparing the inspected object with its spatially aligned ideal reference model. Rotation estimation is crucial for the inspection of rotationally symmetric objects where mechanical manipulation is unable to ensure the correct object rotation. We propose a novel method for in-plane rotation estimation. Rotation is estimated with an ensemble of nearest-neighbor estimators. Each estimator contains a spatially local representation of an object in a feature space for all rotation angles and is constructed with a semi-supervised self-training approach from a set of unlabeled training images. An individual representation in a feature space is obtained by calculating the Histograms of Oriented Gradients (HOG) over a spatially local region. Each estimator votes separately for the estimated angle; all votes are weighted and accumulated. The final estimation is the angle with the most votes. The method was evaluated on several datasets of pharmaceutical tablets varying in size, shape, and color. The results show that the proposed method is superior in robustness with comparable speed and accuracy to previously proposed methods for rotation estimation of pharmaceutical tablets. Furthermore, all evaluations were performed with the same set of parameters, which implies that the method requires minimal human intervention. Despite the evaluation focused on pharmaceutical tablets, we consider the method useful for any application that requires robust real-time in-plane rotation estimation.

We propose encoder-centric stepwise models for extractive summarization using structured transformers -HiBERT Extended Transformers (Ainslie et al., 2020). We enable stepwise summarization by injecting the previously generated summary into the structured transformer as an auxiliary sub-structure. Our models are not only efficient in modeling the structure of long inputs, but they also do not rely on task-specific redundancy-aware modeling, making them a general purpose extractive content planner for different tasks. When evaluated on CNN/DailyMail extractive summarization, stepwise models achieve state-of-the-art performance in terms of Rouge without any redundancy aware modeling or sentence filtering. This also holds true for Rotowire tableto-text generation, where our models surpass previously reported metrics for content selection, planning and ordering, highlighting the strength of stepwise modeling. Amongst the two structured transformers we test, stepwise Extended Transformers provides the best performance across both datasets and sets a new standard for these challenges. 1 * Equal contribution. 1 The code and data are available at https://github. com/google-research/google-research/ tree/master/etcsum.

No abstract

No abstract