Deep_Learning_For_Visual_Recognition (Pytorch)

Deep Learning For Visual Recognition (STA9131)

1. KNN

• Implement KNN and use cross-validation to find optimal K

2. Linear_classifier

2-1. SVM loss

2-2. Softmax loss

3. Fully connected neural network

3-1. Two layer net

3-2. Fully connected layer

4. Convolutional networks

Make modules such as Convolutional layer forward, backward, batchnormalization and make deep convolutional networks.

4.1 Deep convolutional networks

• Forward, backward for Convolutional layer, ReLU layer, Pooling layer, BN

4.1 Initialization

• Kaiming initialization

4.1 BN (Batchnormalization)

• BN
• spatial BN

5. VAE

• pytorch autograd and nn

6. GAN

7. Motion captioning with RNNs and LSTMs

7-1. RNNs

7-2. LSTM

• https://dgkim5360.tistory.com/entry/understanding-long-short-term-memory-lstm-kr