A Comprehensive Study of ImageNet Pre-Training for Historical Document Image Analysis
Authors:
Linda Studer,
Michele Alberti,
Vinaychandran Pondenkandath,
Pinar Goktepe,
Thomas Kolonko,
Andreas Fischer,
Marcus Liwicki,
Rolf Ingold
Abstract:
Automatic analysis of scanned historical documents comprises a wide range of image analysis tasks, which are often challenging for machine learning due to a lack of human-annotated learning samples. With the advent of deep neural networks, a promising way to cope with the lack of training data is to pre-train models on images from a different domain and then fine-tune them on historical documents.…
▽ More
Automatic analysis of scanned historical documents comprises a wide range of image analysis tasks, which are often challenging for machine learning due to a lack of human-annotated learning samples. With the advent of deep neural networks, a promising way to cope with the lack of training data is to pre-train models on images from a different domain and then fine-tune them on historical documents. In the current research, a typical example of such cross-domain transfer learning is the use of neural networks that have been pre-trained on the ImageNet database for object recognition. It remains a mostly open question whether or not this pre-training helps to analyse historical documents, which have fundamentally different image properties when compared with ImageNet. In this paper, we present a comprehensive empirical survey on the effect of ImageNet pre-training for diverse historical document analysis tasks, including character recognition, style classification, manuscript dating, semantic segmentation, and content-based retrieval. While we obtain mixed results for semantic segmentation at pixel-level, we observe a clear trend across different network architectures that ImageNet pre-training has a positive effect on classification as well as content-based retrieval.
△ Less
Submitted 22 May, 2019;
originally announced May 2019.
A topology-oblivious routing protocol for NDN-VANETs
Authors:
Eirini Kalogeiton,
Thomas Kolonko,
Torsten Braun
Abstract:
Vehicular Ad Hoc Networks (VANETs) are characterized by intermittent connectivity, which leads to failures of end-to-end paths between nodes. Named Data Networking (NDN) is a network paradigm that deals with such problems, since information is forwarded based on content and not on the location of the hosts. In this work, we propose an enhanced routing protocol of our previous topology-oblivious Mu…
▽ More
Vehicular Ad Hoc Networks (VANETs) are characterized by intermittent connectivity, which leads to failures of end-to-end paths between nodes. Named Data Networking (NDN) is a network paradigm that deals with such problems, since information is forwarded based on content and not on the location of the hosts. In this work, we propose an enhanced routing protocol of our previous topology-oblivious Multihop, Multipath, and Multichannel NDN for VANETs (MMM-VNDN) routing strategy that exploits several paths to achieve more efficient content retrieval. Our new enhanced protocol, i mproved MMM-VNDN (iMMM-VNDN), creates paths between a requester node and a provider by broadcasting Interest messages. When a provider responds with a Data message to a broadcast Interest message, we create unicast routes between nodes, by using the MAC address(es) as the distinct address(es) of each node. iMMM-VNDN extracts and thus creates routes based on the MAC addresses from the strategy layer of an NDN node. Simulation results show that our routing strategy performs better than other state of the art strategies in terms of Interest Satisfaction Rate, while kee** the latency and jitter of messages low.
△ Less
Submitted 19 November, 2018; v1 submitted 27 November, 2017;
originally announced November 2017.
