CS 180 Project 5a: The Power of Diffusion Models!
By Shreyas Prasanna
In part A we will play around with diffusion models, implement diffusion sampling loops, and use them for other tasks such as inpainting and creating optical illusions.
Part 0
Random Seed = 222
Output for the 3 Text Prompts
The quality of these images with respect to their propmts is very good.
Part 1
Output for test image at noise level [250, 500, 750]
Output for level 250
Output for level 500
Output for level 750
Below are the results for Iterative Denoising
Now, the Diffusion Model Sample (5 of them) are given below
Results(samples 1, 2, 3, 4, 5):
High Quality Image and Low Quality Image Results(samples 1, 2, 3, 4, 5):
The edited images at noise levels [1, 3, 5, 7, 10, 20] are below:
Editing Hand-Drawn and Web Images
Below are the images used
Below are the Visual Anagrams Produced:
Old man(people around campfire) , Red Panda (kitchen table) , Horse (snowy mountains)
Below are the results of producing a couple of Hybrid Images:
**Please note that my Inpaint and edited image results were for some reason not uploaded on the website. I keep getting an unknown error. I have implemented the correct code.**
CS 180 Project 5b: Diffusion Models from Scratch!
Below is the information i got from training and implementing the model
Below is the result of training the model
Here is the Training Loss Curve we obtain
Below are the results of running the model on our Data
**Please note that my image results could not be uploaded for a similar reason as part a. I fail to understand why this happened, but I have implemented the correct code.