Few-Shot Learning for Online Large Scale Fine-Grained Image Recognition
Тип публікації :
Магістерська робота
Дата випуску :
2021
Автор(и) :
Holubakha Mykyta Ihorovich
Мова основного тексту :
Англійська
eKNUTSHIR URL :
Цитування :
[APA 7] Holubakha, M. I. (2021). Few-Shot Learning for Online Large Scale Fine-Grained Image Recognition. [Master's thesis, Київський національний університет імені Тараса Шевченка]. eKNUTSHIR. https://ir.library.knu.ua/handle/123456789/2977
[ДСТУ] Holubakha M. I. Few-Shot Learning for Online Large Scale Fine-Grained Image Recognition : кваліфікаційна робота магістра : 12 Інформаційні технології. Kyiv, 2021. 58 p. URL: https://ir.library.knu.ua/handle/123456789/2977 (date of access: 17.07.2026).
In this work we have analyzed the existing body of research on few-shot learning and the various methods used to solve the few-shot image classification problem.
We have treated, both theoretically and experimentally, embedding learning methods, multitask methods, different kinds of metric losses. Tried various self- and semi-supervised learning methods like MoCo, SwAV, SimCLR, BYOL and SimSiam.
In addition, we have tackled the problem of performing online large-slace image
recognition. It requires solving tasks like Out-Of-Distribution detection and clusterization, class discovery.
As a result of this work, we have proposed a mechanism for implementing a
system to solve the problem of fine-grained large-scale image recognition using
few-shot learning methods.
We have conducted thourough qualitative and quantitative analysis on the application of various few-shot learning methods for image classification implemented in PyTorch both in benchmark and real-world data.
This methods will further be improved on and tested on real-world product data.
We have treated, both theoretically and experimentally, embedding learning methods, multitask methods, different kinds of metric losses. Tried various self- and semi-supervised learning methods like MoCo, SwAV, SimCLR, BYOL and SimSiam.
In addition, we have tackled the problem of performing online large-slace image
recognition. It requires solving tasks like Out-Of-Distribution detection and clusterization, class discovery.
As a result of this work, we have proposed a mechanism for implementing a
system to solve the problem of fine-grained large-scale image recognition using
few-shot learning methods.
We have conducted thourough qualitative and quantitative analysis on the application of various few-shot learning methods for image classification implemented in PyTorch both in benchmark and real-world data.
This methods will further be improved on and tested on real-world product data.
Галузі знань та спеціальності :
12 Інформаційні технології
122 Комп’ютерні науки
Файл(и) :![Ескіз]()
Вантажиться...
Формат :
Adobe PDF
Розмір :
2.09 MB
Контрольна сума :
(MD5):e6642845b5434fa125b2cfd223a0acf6
Якщо не вказано інше, ця робота розповсюджується на умовах ліцензії Creative Commons Attribution-NonCommercial 4.0 International