Harpreet Sahota

Seriously, Memes Are All We Need Harpreet Sahota · Follow Published in Voxel51 · 10 min read · -- A dataset of memes in FiftyOne format The idea for this project occurred when the Janus Pro model by DeepSeek AI dropped. I was reading through the paper, and in my hyped frame of mind, I thought I read the model excels in MEME Perception…in fact, what the paper actually said was the model excels in MME Perception.

Exploring the Intersection of Vision Language Models and Audio Data Harpreet Sahota · Follow Published in Voxel51 · 13 min read · -- ESC-10 dataset parsed as spectrograms into a FiftyOne Dataset I recently came across a paper that made me wonder: could we actually use vision language models to understand audio? The paper, Vision Language Models Are Few-Shot Audio Spectrogram Classifiers, introduces and explores Visual Spectrogram Classification (VSC).

Testing Vision Model Robustness: A hands-on tutorial for evaluating vision model performance on synthetic data using embedding analysis and zero-shot classification Harpreet Sahota · Follow Published in Voxel51 · 11 min read · Feb 12, 2025 -- Exploring ImageNet-D in FiftyOne ImageNet-D is a new benchmark of synthetically generated images (via Stable Diffusion) that’s pushing image classification models to their breaking points with challenging images and revealing critical failures in model...