An exploration of the DeepFloyd IF diffusion model, detailing image generation from text prompts and various denoising techniques to enhance image quality.
About:
Vijay Kethanaboyina's personal website
Website:
Specializations:
Outgoing Links:
Subscribe to RSS:
This post outlines the implementation and comparison of MSE and Flow Matching policies for training an agent in a 2D manipulation task, emphasizing the superior performance of Flow Matching.
The blog post details the author's implementation of an image colorization algorithm as part of a computer vision class at UC Berkeley. It describes the process of colorizing grayscale images from the Prokudin-Gorskii photo collec...
The blog post discusses an assignment from the CS 182: Deep Neural Networks course at UC Berkeley, where the author created an interactive diagram to visualize the Qwen3 Mixture-of-Experts (MoE) architecture. It includes compariso...
The blog post details the author's experience implementing Neural Radiance Fields (NeRFs) from scratch as part of a computer vision class at UC Berkeley. It covers the entire process, including camera calibration, dataset collecti...
The blog post details a project for CS 180 at UC Berkeley, focusing on image processing techniques using filters. It covers the implementation of 2D convolution operations in NumPy, comparing different methods for efficiency. The ...
6.2Random Research Ideas
2024-05-25 • research linguistics epistemology
The blog post explores various experimental ideas and questions related to language models, artificial intelligence, and linguistics. It discusses the potential of a robot learning to walk through random code mutations, methods fo...
The blog post discusses the creation of an interactive language-learning app that uses large language models (LLMs) to generate 'cognateful' sentences in French, aimed at English speakers. It explains the concept of comprehensible...
The post introduces VJScript, a new writing system designed to address the inadequacies of English spelling. It explains the rationale behind creating a new system, highlighting historical attempts at spelling reform. VJScript sep...
The blog post details a project for a computer vision course at UC Berkeley, focusing on image warping, mosaic generation, and corner detection algorithms. It outlines the steps taken to capture images, recover homographies, and a...
The blog post summarizes insights from a speaker event featuring Tyler Bosmeny, founder of Clever, an ed-tech company. Key points include Clever's monetization model, the importance of domain expertise in product development, chal...
The author discusses their motivation for starting a personal website, emphasizing the importance of independent online presence amidst the rise of 'web feudalism' where content is hosted on larger platforms. They reflect on their...