Back To Projects
Brightness helps CNN classify a subset of the images from Google Quick Draw
Serena F.
We made a CNN that learned to recognize and classify sketches from Google’s Quick Draw dataset and implemented a novel brightness feature to test if it increased the accuracy.
We made a CNN that learned to recognize and classify sketches from Google’s Quick Draw dataset and implemented a novel brightness feature to test if it increased the accuracy. The results displayed up to a 4% increase in validation accuracy. To conduct experiments, we created 3 models, 1 with brightness and two without. In addition, we used different numbers of epochs, metalayers, kernels, and sizes of kernels and dropouts.
Serena F.
Clayton Greenberg
Teaching Professor from UPenn
Related Projects
Classifying AI-Generated Music with AI Models
The importance of this research is twofold: addressing copyright issues and fostering the adoption of AI in the music industry. As AI-generated music blurs the line between human and AI creation, concerns regarding copyright ownership and artistic attribution become paramount [2]. To address these challenges, we created an AI model to differentiate AI-generated music from human-composed music, specifically focusing on music generated by the JukeBox model.
Siddharth M.
Mentored by Anthony Cuturrufo
Predicting the Presence of Autism Spectrum Disorder Based on Eye-Tracking Scan Path Images
This research explores using eye-tracking data and machine learning to predict Autism Spectrum Disorder (ASD), achieving 74.6% accuracy with logistic regression, highlighting eye-tracking's potential for early, non-invasive ASD diagnosis and its applicability to other neurological conditions.
Sara C.
Mentored by Emily Broadhurst
Self-supervised DINOv2 versus Supervised CNNs for Insurance Claim Cost Predictions
This study investigates whether vehicle insurance claim amounts can be predicted directly from images of damaged cars by comparing CNN models, the self-supervised DINOv2 model, and GPT-5.5 as a vision language model. Results showed that DINOv2 performed best, achieving an R² of 0.12 compared with 0.07 for ResNet152V2 and 0.08 for ResNeXt50, while GPT-5.5 tended to overestimate repair costs, particularly for severe damage.
Rohan V.
Mentored by Ivan Felipe Rodriguez
© 2022 Inspirit AI. All Rights Reserved.