Computer Vision

HumorDB is a novel image-only dataset designed to advance visual humor understanding in AI systems. It consists of carefully curated image pairs with contrasting humor ratings, emphasizing subtle visual cues that trigger humor while mitigating potential biases. The dataset enables evaluation through binary classification, range regression, and pairwise comparison tasks. Authors: Vedaant Jain, Felipe Feitosa, Gabriel Kreiman

Jun 12, 2024

HumorDB is a novel dataset for benchmarking and advancing visual humor understanding in AI systems, consisting of curated image pairs with contrasting humor ratings and enabling various evaluation tasks.

Jun 7, 2024

This project addresses limitations in Parkinson’s Disease (PD) research by creating synthetic data that aims to simulate changes in facial features associated with PD progression. Using diffusion models and inpainting techniques, we developed a pipeline to generate realistic image pairs representing the transition from healthy to PD-affected facial states. Additionally, we utilized evaluations based on training classification models on the synthetic data. We also explored Model generalization using subset of FFHQ dataset and saw improvement by 5% over previous model baseline.

May 12, 2024

This project addresses the challenge of extracting structured information from academic resumes, which often span multiple pages and contain complex, domain-specific content. We developed a novel approach combining document layout analysis and sequence tagging to accurately segment and extract key information from various resume sections.

May 10, 2023

This project implements artistic style transfer using Convolutional Neural Networks (CNNs) in Rust. It combines the content of one image with the artistic style of another, creating unique visual outputs. The system is deployed as a web application, allowing users to easily interact with the model through a REST API.

Dec 10, 2022