Bilgisayar Mühendisliği Bölümü Koleksiyonu
Permanent URI for this collectionhttps://hdl.handle.net/20.500.11779/1940
Browse
4 results
Search Results
Article Citation - WoS: 3Citation - Scopus: 3A Benchmark Dataset for Turkish Data-To Generation(Elsevier, 2023-01-01) Demir, Şeniz; Öktem, SezaIn the last decades, data-to-text (D2T) systems that directly learn from data have gained a lot of attention in natural language generation. These systems need data with high quality and large volume, but unfortunately some natural languages suffer from the lack of readily available generation datasets. This article describes our efforts to create a new Turkish dataset (Tr-D2T) that consists of meaning representation and reference sentence pairs without fine-grained word alignments. We utilize Turkish web resources and existing datasets in other languages for producing meaning representations and collect reference sentences by crowdsourcing native speakers. We particularly focus on the generation of single-sentence biographies and dining venue descriptions. In order to motivate future Turkish D2T studies, we present detailed benchmarking results of different sequence-to-sequence neural models trained on this dataset. To the best of our knowledge, this work is the first of its kind that provides preliminary findings and lessons learned from the creation of a new Turkish D2T dataset. Moreover, our work is the first extensive study that presents generation performances of transformer and recurrent neural network models from meaning representations in this morphologically-rich language.Article Citation - WoS: 5Citation - Scopus: 7Founsure 1.0: an Erasure Code Library With Efficient Repair and Update Features(Elsevier, 2021-01-01) Arslan, Şuayb ŞefikFounsure is an open-source software library that implements a multi-dimensional graph-based erasure coding entirely based on fast exclusive OR (XOR) logic. Its implementation utilizes compiler optimizations and multi-threading to generate the right assembly code for the given multi-core CPU architecture with vector processing capabilities. Founsure possesses important features that shall find various applications in modern data storage, communication, and networked computer systems, in which the data needs protection against device, hardware, and node failures. As data size reached unprecedented levels, these systems have become hungry for network bandwidth, computational resources, and average consumed power. To address that, the proposed library provides a three-dimensional design space that trades off the computational complexity, coding overhead, and data/node repair bandwidth to meet different requirements of modern distributed data storage and processing systems. Founsure library enables efficient encoding, decoding, repairs/rebuilds, and updates while all the required data storage and computations are distributed across the network nodes.Article Citation - WoS: 30Citation - Scopus: 44An Efficient Framework for Visible-Infrared Cross Modality Person Re-Identification(Elsevier, 2020-09-01) Gökmen, Muhittin; Başaran, Emrah; Kamasak, Mustafa E.Visible-infrared cross-modality person re-identification (VI-ReId) is an essential task for video surveillance in poorly illuminated or dark environments. Despite many recent studies on person re-identification in the visible domain (ReId), there are few studies dealing specifically with VI-ReId. Besides challenges that are common for both ReId and VI-ReId such as pose/illumination variations, background clutter and occlusion, VI-ReId has additional challenges as color information is not available in infrared images. As a result, the performance of VI-ReId systems is typically lower than that of ReId systems. In this work, we propose a four-stream framework to improve VI-ReId performance. We train a separate deep convolutional neural network in each stream using different representations of input images. We expect that different and complementary features can be learned from each stream. In our framework, grayscale and infrared input images are used to train the ResNet in the first stream. In the second stream, RGB and three-channel infrared images (created by repeating the infrared channel) are used. In the remaining two streams, we use local pattern maps as input images. These maps are generated utilizing local Zernike moments transformation. Local pattern maps are obtained from grayscale and infrared images in the third stream and from RGB and three-channel infrared images in the last stream. We improve the performance of the proposed framework by employing a re-ranking algorithm for post-processing. Our results indicate that the proposed framework outperforms current state-of-the-art with a large margin by improving Rank-1/mAP by 29.79%/30.91% on SYSU-MM01 dataset, and by 9.73%/16.36% on RegDB dataset.Article Citation - WoS: 13Citation - Scopus: 16Face Recognition With Patch-Based Local Walsh Transform(Elsevier, 2018-02-01) Uzun-Per, Meryem; Gökmen, MuhittinIn this paper, we present a novel dense local image representation method called Local Walsh Transform (LWT)by applying the well-known Walsh Transform (WT) to each pixel of an image. The LWT decomposes an image into multiple components, and produces LWT complex images by using the symmetrical relationship between them. Cascaded LWT (CLWT) is also a dense local image representation obtained by applying the LWT again to real and imaginary parts of LWT complex images. Applying the LWT once more to real and imaginary parts of LWT complex images increases the success rate especially on low resolution images. In order to combine the advantages of sparse and dense local image representations, we present Patch-based LWT (PLWT) and Patch-based CLWT (PCLWT) by applying the LWT and CLWT, respectively, to patches extracted around landmarks of multi-scaled face images. The extracted high dimensional features of the patches are reduced through the application of the Whitened Principal Component Analysis (WPCA). Experimental results show that both thePLWT and PCLWT are robust to illumination and expression changes, occlusion and low resolution. The state-of-the-art performance is achieved on the FERET and SCface databases, and the second best unsupervised category result is achieved on the LFW database.