Бидний тухай
Багш ажилтан
This paper explores in detail the methods, techniques, and technologies employed to capture a high-definition gigapixel panoramic image of Ulaanbaatar, the capital city of Mongolia. Using specialized equipment including a Canon 6D Mark II camera, a Sigma 50-500mm lens, and a Benro Polaris BR209 panoramic head, a series of 1260 photographs were taken, each with a resolution of 3120 x 2080 pixels. These images were then expertly stitched and edited to produce a seamless panoramic display. We further delve into the complexities of storing such a large image file and the software solutions for presenting this on a web platform
In this paper, we presented two POS taggers for Mongolian, namely Neural Networks - Multilayer Perceptron and Hidden Markov Model with Viterbi. The accuracy of the former tagger is 95.6%, whereas the latter tagger is 85.6%. Also, we compared the performance of our taggers with the previous works. The comparison shows that the Neural Network tagger performs better for Mongolian POS tagging than other approaches. Our dataset consists of about 5000 sentences and includes almost 100,000 words for training and testing.
Компьютерын хэл шинжлэл нь дэлхийн хэл бүрд харилцан адилгүй хөгжсөөр байна. Аливаа хэлний хувьд хэл боловсруулалтын судалгаа, арга, техникүүд хөгжих нь мэдээллийн технологийнх нь салбарт чухал ач холбогдолтой. Хэдийгээр манай Монгол хэл боловсруулалт харьцангуй хөгжиж байгаа хэдий ч зарим нэг чухал судалгаанууд хийгдээгүй хэвээр байна. Учир нь зарим тохиолдолд эдгээр судалгааг хийх нөөц ба хэрэгслүүд хараахан бүрдээгүйд оршино. Эдгээр судалгаануудын нэг нь Нээлттэй Мэдээлэл Задлал юм. Энэхүү өгүүллийн зорилго нь ийм системийг монгол хэл дээр байгуулах боломж, нөхцөл бүрдсэн эсийг судалж ашиглагдаж болох аргуудыг танилцуулах юм.
As virtual tour and virtual reality technology advances, resolution with gigapixel panoramic images become available. Visualization technique for gigapixel panoramic image needs to perform fast without any delay to access gigapixel ultra high resolution image data through wired or wireless internet by a Mobile, Tablet or a standard PC. Also, several popular visualization formats like equirectangular projection were developed to map a real scene to a panoramic image. However, Equirectangular projection may not the best choice when considering an image quality because of the distortion on both poles of the sphere. In this paper, we show the result of the evaluation for the projection formats using various comparison method to find which projection format (Equirectangular Projection, Cubemap Projection and Octahedron projection) and visualization technologies provides better image quality and speed of loading over internet. We exclude other format that has lower quality than the equirectangular projection format used in this experiment.
The aim of this paper is to describe a a simple Neural Networks model - Multilayer Perceptron for Mongolian Part-of Speech tagging. We used about 1400 manually tagged sentences for training and testing from Mongolian Penn Treebank. The performance of the model is 80.78% which we consider a promising result. Also another contribution of this work is that we make our testing data online for the sake of the development of Mongolian Part-of Speech tagging research.
In this paper, we describe MongoIE, an Open Information Extraction(OpenIE) system for the Mongolian language. We present the characteristic of the language and, after analyzing the available preprocessing tools, we describe the features used for building the system. We have implemented two different approaches: 1)Rule-based and 2)Classification. Here, we describe them, analyze their errors and present their results. In the best of our knowledge, this is the first attempt in building Open IE systems for Mongolian. We conclude by suggesting possible future improvements and directions.
In this paper, we describe MongoIE, an Open Information Extraction(OpenIE) system for the Mongolian language. We present the characteristic of the language and, after analyzing the available preprocessing tools, we describe the features used for building the system. We have implemented two different approaches: 1)Rule-based and 2)Classification. Here, we describe them, analyze their errors and present their results. In the best of our knowledge, this is the first attempt in building Open IE systems for Mongolian. We conclude by suggesting possible future improvements and directions.
Энэхүү үзүүлэн илтгэл нь хөдөлгөөн илрүүлэгчийн хэрэглээ болон ерөнхий ойлголтын талаар, түүнчлэн 3-хэмжээст моделийг хөдөлгөхөд зориулан хөдөлгөөн илрүүлэгчээс гарган авсан өгөгдлүүдэд суурилсан туршилтыг танилцуулах болно. Туршилтыг сүүлийн үеийн сенсор технологи болох “Shadow” хөдөлгөөн баригч системийг ашиглан хүний хөдөлгөөний хөдөлгөөний мэдээлэл дээр тулгуурлан хийсэн бөгөөд уг мэдээллээ ашиглан 3 хэмжээст моделийн хөдөлгөөнийг удирдахад ашигладаг инерцийн сенсор аргуудыг (Гироскоп, Хурдасгалыг хэмжигч хэрэгслүүд, г.м) туршиж хэрэгжүүллээ.
