|
|
Facial image restoration
Bako, Matúš ; Herout, Adam (referee) ; Hradiš, Michal (advisor)
In this thesis, I tackle the problem of facial image super-resolution using convolutional neural networks with focus on preserving identity. I propose a method consisting of DPNet architecture and training algorithm based on state-of-the-art super-resolution solutions. The model of DPNet architecture is trained on Flickr-Faces-HQ dataset, where I achieve SSIM value 0.856 while expanding the image to four times the size. Residual channel attention network, which is one of the best and latest architectures, achieves SSIM value 0.858. While training models using adversarial loss, I encountered problems with artifacts. I experiment with various methods trying to remove appearing artefacts, which weren't successful so far. To compare quality assessment with human perception, I acquired image sequences sorted by percieved quality. Results show, that quality of proposed neural network trained using absolute loss approaches state-of-the-art methods.
|
|
|
Playing Gomoku with Neural Networks
Bako, Matúš ; Kolář, Martin (referee) ; Hradiš, Michal (advisor)
The goal of this thesis is to create an artificial intelligence for playing Gomoku. While conventional methods usually use state space search combined with predefined rules, this artificial intelligence uses state space search and learned neural networks. A strategic network computes probability distribution for given a board state and a value network determines outcome of the game from a given board state. I trained multiple architectures of neural networks with different number of convolutional layers and different sizes of convolution kernels. Experiments show, that it is problematic to end a game without using the value network or search algorithm, but the strategic network can be used as a heuristic for choosing next move. Despite using relatively small dataset, created artificial intelligence is capable of beating weaker programs from Gomocup competition.
|
|
|
Facial image restoration
Bako, Matúš ; Herout, Adam (referee) ; Hradiš, Michal (advisor)
In this thesis, I tackle the problem of facial image super-resolution using convolutional neural networks with focus on preserving identity. I propose a method consisting of DPNet architecture and training algorithm based on state-of-the-art super-resolution solutions. The model of DPNet architecture is trained on Flickr-Faces-HQ dataset, where I achieve SSIM value 0.856 while expanding the image to four times the size. Residual channel attention network, which is one of the best and latest architectures, achieves SSIM value 0.858. While training models using adversarial loss, I encountered problems with artifacts. I experiment with various methods trying to remove appearing artefacts, which weren't successful so far. To compare quality assessment with human perception, I acquired image sequences sorted by percieved quality. Results show, that quality of proposed neural network trained using absolute loss approaches state-of-the-art methods.
|
|
|
Playing Gomoku with Neural Networks
Bako, Matúš ; Kolář, Martin (referee) ; Hradiš, Michal (advisor)
The goal of this thesis is to create an artificial intelligence for playing Gomoku. While conventional methods usually use state space search combined with predefined rules, this artificial intelligence uses state space search and learned neural networks. A strategic network computes probability distribution for given a board state and a value network determines outcome of the game from a given board state. I trained multiple architectures of neural networks with different number of convolutional layers and different sizes of convolution kernels. Experiments show, that it is problematic to end a game without using the value network or search algorithm, but the strategic network can be used as a heuristic for choosing next move. Despite using relatively small dataset, created artificial intelligence is capable of beating weaker programs from Gomocup competition.
|
guest ::
