Description
Contains two task Task1: latent space interaction with VAE Task2: masked image reconstruction with VAE
Unformatted Attachment Preview
Assignment Objectives:
The primary goal of this assignment is to implement a Variational Autoencoder (VAE). The assignment comprises two tasks:
Task 1: Latent Space Interaction with VAE
• Build and train a Variational Autoencoder (VAE) model to learn a 6 dimensional latent space representation of a facial data set.
• Interact with the decoder part of the trained VAE by creating a graphical user interface (GUI) with six sliders, representing the values in a
• Reconstruct and display facial images in real time when user interacts with the sliders using a mouse. .
Task 2: Masked Image Reconstruction with VAE
• Use the same data set as in Task 1 to train a second VAE to handle masked facial images. The mask is assumed to be a square with variab
• Once the training is done, enable the user to load an image and interactively change the position of a square mask over the selected image
using the trained VAE from task 1 and also display the reconstructed image using the trained VAE from task 2.
• The masked portion of the image should be set to all zeros.
• The normalized size of the mask should be adjustable from 0 to 0.5. The normalized position of the top left of the mask should be betwee
• Reconstruct the image in real-time as the user moves and changes the mask’s size, displaying the associated Mean Squared Error (MSE).
Datasets:
You have the option to use any of the following facial data sets:
• LFW (Labeled Faces in the Wild): Contains approximately 13,000 labeled facial images.
• CelebA: Comprises over 200,000 celebrity images.
• FER2013 (Facial Expression Recognition 2013): Includes around 35,000 images for facial expression analysis.
• IMDB-WIKI: Contains over half a million images of celebrities.
• CASIA WebFace: Includes over 500,000 images of celebrities.
• MS Celeb 1M: Consists of around 10 million images of celebrities.
• 300 Faces In-the-Wild (300W): A dataset with 68,000 labeled faces displaying varying poses, expressions, and occlusions, often used for
• Multi-PIE: A dataset with more than 750,000 images of 337 individuals, captured under different illumination, pose, and expression cond
• AFLW (Annotated Facial Landmarks in the Wild): Contains over 25,000 in-the-wild facial images with annotated facial landmarks.
Grading Criteria:
This assignment is entirely optional, and there are no partial credits. To receive a grade for this assignment, all components in Task 1 and Task 2 m
and real-time reconstruction work seamlessly for both tasks.
Notes:
• Your GUI must exactly match the image shown below.
• Apart from the number of latent variables, your are free to choose the architecture of your VAEs.
• Submit your saved trained model with your submission.
• Your program must automatically load the saved model when it starts to run.
• Your program must automatically show the GUI when it runs.
• Do not submit the facial dataset.
• Please ensure that you consult the specific data set’s documentation and comply with usage terms and permissions when working with fac
Submission Guidelines:
•
The first four lines of your submitted files must have the following format:
# Your name (last-name, first-name)
# Your student ID (100x_xxx_xxx)
# Date of submission (yyyy_mm_dd)
# Assignment_nn_kk
• Create a directory and name it according to the submission guidelines and include your files in that directory.
• Zip the directory and upload it to Canvas according to the submission guidelines.
Purchase answer to see full
attachment
