Global pooling layers are an essential part of Convolutional Neural Networks (CNN). They are used to aggregate activations of spatial locations to produce a fixed-size vector in several state-of-the-art CNNs. Global average pooling or global max pooling are commonly used for converting convolutional features of variable size images to a fix-sized embedding. However, both pooling layer types are computed spatially independent: each individual activation map is pooled and thus activations of different locations are pooled together. In contrast, we propose Deep Generalized Max Pooling that balances the contribution of all activations of a spatially coherent region by re-weighting all descriptors so that the impact of frequent and rare ones is equalized. We show that this layer is superior to both average and max pooling on the classification of Latin medieval manuscripts (CLAMM'16, CLAMM'17), as well as writer identification (Historical-WI'17).

Current optical character recognition (OCR) systems commonly make use of recurrent neural networks (RNN) that process whole text lines. Such systems avoid the task of character segmentation necessary for character-based approaches. A disadvantage of this approach is a need of a large amount of annotated data. This can be solved by sing generated synthetic data instead of costly manually annotated ones. Unfortunately, such data is often not suitable for historical documents particularly for quality reasons. This work presents a hybrid approach for generating annotated data for OCR at a low cost. We first collect a small dataset of isolated characters from historical document images. Then, we generate historical looking text lines from the generated characters. Another contribution lies in the design and implementation of an OCR system based on a convolutional-LSTM network. We first pre-train this system on hybrid data. Afterwards, the network is fine-tuned with real printed text lines. We demonstrate that this training strategy is efficient for obtaining state-of-the-art results. We also show that the score of the proposed system is comparable or even better in comparison to several state-of-the-art systems.