2019 CSE deep learning course
This markdown is for Hanyang University Deep Learning course. For more detailed information, please read reports below.
Goal
Build a 2-layered neural network for logistic regression.
Implement the network using both numpy & tensorflow, and compare the results.
Detail
- Generate 128 2-dim vectors between -1 and 1.
- Train using Gradient descent.
- Repeat for 5000 times.
- For more detailed report, read lab1.pdf
Goal
Implement image super-resolution using CNN(SRCNN) with tensorflow.
Detail
- Made every sample pictures into half-size & grayscale.
- Trained cropped samples(32*32) with SRCNN, and compared the results with original pictures.
- For more detailed report, read Lab2_report.pdf
Result
Goal
Implement SRCNN using vanilla RNN with tensorflow.
Detail
- Made every sample pictures into half-size & grayscale.
- Trained cropped samples(32*32) with SRCNN, and compared the results with original pictures.
- Used ReLU for activation function, and needed to calculate 'h' for RNN.
- For more detailed report, read Lab4_report.pdf
Result
-
Learning a Deep Convolutional Network for Image Super-Resolution (ECCV 2014), C Dong et al.
-
Image Super-Resolution Using Deep Convolutional Networks (TPAMI 2015), C Dong et al.