DLND_deep_learning_nanodegree

pytorch-udacity-projects

The notebooks contain the projects of PyTorch Scholarship Challenge and Deep Learning Nanodegree from Udacity and facebook.

Project 1: Flower image classification

The flower_image_classification.ipynb contains the project Flower image classifier. The 102 Category Flower Dataset from Visual Geometry Group, University of Oxford, is used.

The following steps are described:

• Preprocessing
• Transfer learning
• Saving and loading model checkpoint
• Inference and Validation

Analysis:

• Model used: ResNet101
• Epochs trained: 20
• Validation accuracy: 94.13%
• Optimizer used: Adam
• Loss used: CrossEntropyLoss
• Scheduler used: StepLR
• Device used: cuda
• Comments: fc layer replacement with combination of linear layers with Dropout regularization

Project 2: Dog breed classification

The dog_breed_classification.ipynb contains the project Dog-Breed Classifier.

The following tasks were completed:

• Detect Humans
• Detect Dogs
• Create a CNN to Classify Dog Breeds (from Scratch)
• Create a CNN to Classify Dog Breeds (using Transfer Learning)

Project 3: TV Script generation

The tv_script_generation.ipynb contains the project TV Script generation.

• The project uses LSTM.

jerry: sittin' watches.
sales woman: oh yeah, that's the one that was.
captain: oh yeah.
elaine: oh, hi.
officer: oh.
elaine: hi.
manager: hi, woah, woah.
sales woman: hi, let's go.
sales woman: alright, let's get it.
jerry:(to george) i don't want to go back to david peterman! i can't believe it!
jerry: hey- what's happening with you!
kramer: hey!
glenn:(to elaine) hey, jerry.
elaine: oh, that's right, i'm gonna get out. i'm starving, and you want to get out of the coma?
captain: well, it's like a liar.
clerk: well, i don't think so.
kramer: well, i think i'd like to be honest with this plane.
elaine:(pointing to jerry) well, it's a lot of pressure.
sales woman:(confused) i think you're a human dumb.
george: well, i think i'm going to have any differently.
sales woman: what do you care about it?
george: yeah, yeah.
elaine: yeah.
helen: i can't find the cooks.
george: well, i think i could be able to be able to be honest in the toilet.
jerry: what is this doorman?
officer: yes.
george: i can't believe this type!
jerry: what? what?
captain: oh, yeah.
sales woman: i am aware.
elaine: well, i don't think i have a problem with a movie.
jerry: oh, my god!
sales woman: hey, jerry?
officer: hey, what's going on here?
sales woman: well, i was terrified- 9 vest.
elaine:(to jerry) i don't think so.
kramer: oh.
mr. ross: i can't.

Project 4: Face Generation using DCGAN

The face_generation_dcgan.ipynb contains the project on face generation.

• Model used: DCGAN (Deep Convolutional Generative Adversarial Networks)
  • Discriminator: strided convolution > batch norm > leaky ReLU
  • Generator: transpose convolution > batch norm > ReLU
• Optimizers: Adam
• learning rate: 0.0002
• beta1 = 0.5 (changed from 0.9)
• beta2 = 0.999 (default value)

Project 5: Deploying sentiment analysis model on Sagemaker

The sagemaker_sentiment_analysis.ipynb contains the project Deploying sentiment analysis model in PyTorch on AWS using Amazon Sagemaker.

• Download or otherwise retrieve the data.
• Process / Prepare the data.
• Upload the processed data to S3.
• Train a chosen model.
• Test the trained model (typically using a batch transform job).
• Deploy the trained model.
• Use the deployed model for web app.
  • Setting up a Lambda function.
  • Setting up API Gateway.

Project 6: Neural Style Transfer

The style_transfer.ipynb decribes the style transfer implementation of Image Style Transfer Using Convolutional Neural Networks by Leon A. Gatys et al.

The Abstract of the paper is:

Rendering the semantic content of an image in different styles is a difficult image processing task. Arguably, a major limiting factor for previous approaches has been the lack of image representations that explicitly represent semantic information and, thus, allow to separate image content from style. Here we use image representations derived from Convolutional Neural Networks optimised for object recognition, which make high level image information explicit. We introduce A Neural Algorithm of Artistic Style that can separate and recombine the image content and style of natural images. The algorithm allows us to produce new images of high perceptual quality that combine the content of an arbitrary photograph with the appearance of numerous well-known artworks. Our results provide new insights into the deep image representations learned by Convolutional Neural Networks and demonstrate their potential for high level image synthesis and manipulation.

You can read the summary of the above paper here.

content image	style image	output image