Image generation by back-propagation on input using a discriminator network

Download

10424142.pdf

Date

2021-9-08

Author

Taplı, Merve

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

329
views

243
downloads

In this thesis, we propose an image generation method that only involves a discriminator network; no generator or decoder networks are required. To generate an image, we iteratively apply an adversarial attack on the discriminator by updating the input image, which is noise at the beginning, to maximize the discriminator's output score. Generated images are then used as negative examples, together with the real images as positive examples, to fine-tune the discriminator. After several rounds of generation and fine-tuning, the generated images start to look real. To show the effectiveness of our method, we present promising results on MNIST, Yale Face, and EMNIST datasets. On MNIST, our FID score (28.8) is comparable to those of the state-of-the-art GANs.

Subject Keywords

Computer vision, Image generation, Adversarial training, Convolutional neural network

URI

https://hdl.handle.net/11511/93100

Collections

Graduate School of Natural and Applied Sciences, Thesis

Suggestions

OpenMETU
Core

Image generation using only a discriminator network with gradient norm penalty Yeşilçimen, Cansu Cemre; Akbaş, Emre; Department of Computer Engineering (2022-9) This thesis explores the idea of generating images using only a discriminator network by extending a previously proposed method (Tapli, 2021) in several ways. The base method works by iteratively updating the input image, which is pure noise at the beginning while increasing the discriminator's score. We extend the training procedure of the base network by adding the following new losses: (i) total variation, (ii) N-way classification (if labels are available), and (iii) gradient norm penalty on real exam...
KINSHIPGAN: SYNTHESIZING OF KINSHIP FACES FROM FAMILY PHOTOS BY REGULARIZING A DEEP FACE NETWORK Ozkan, Savas; Ozkan, Akin (2018-10-10) In this paper, we propose a kinship generator network that can synthesize a possible child face by analyzing his/her parent's photo. For this purpose, we focus on to handle the scarcity of kinship datasets throughout the paper by proposing novel solutions in particular. To extract robust features, we integrate a pre-trained face model to the kinship face generator. Moreover, the generator network is regularized with an additional face dataset and adversarial loss to decrease the overfitting of the limited s...
Shape : representation, description, similarity and recognition Arıca, Nafiz; Yarman Vural, Fatoş Tunay; Department of Computer Engineering (2003) In this thesis, we study the shape analysis problem and propose new methods for shape description, similarity and recognition. Firstly, we introduce a new shape descriptor in a two-step method. In the first step, the 2-D shape information is mapped into a set of 1-D functions. The mapping is based on the beams, which are originated from a boundary point, connecting that point with the rest of the points on the boundary. At each point, the angle between a pair of beams is taken as a random variable to define...
Change detection in aerial images Borchani, M; Cloppet, F; Atalay, Mehmet Volkan; Stamon, G (2004-01-01) This paper deals with how to characterize texture and how to get a good description of images with a minimal number of parameters. This procedure is more objective than textual data. Texture characterization has been used in a matching system to detect changes in couples of aerial images taken at two different times using different order of statistics to describe images. The results are quite encouraging.
Direction Adaptive Super-Resolution Imaging Turgay, Emre; Akar, Gözde (2009-04-11) In this paper a novel edge-presenting super-resolution (SR) image reconstruction method is proposed. The proposed maximum a-posteriori (MAP) based estimator uses gradient direction and amount for optimal noise reduction while presenting the edges. Compared to the other edge-presenting methods, the proposed algorithm uses the gradient direction for optimum regularization. The proposed method estimates gradient amplitude and direction at each iteration. This gradient map guides the SR reconstruction stage thr...

Citation Formats

M. Taplı, “Image generation by back-propagation on input using a discriminator network,” M.S. - Master of Science, Middle East Technical University, 2021.