Voxel51

?? Voxel51 co-founder Jason Corso weighs in on safe AI with Miranda Nazzaro in?The Hill. Check out the full article: https://lnkd.in/e9_74AKm #AISafety #SafeAI #TechIndustry #TechInnovation #Innovation

Calling all Connecticut AI builders! The Visual AI Hackathon is coming to the University of Connecticut! ?? Got a laptop, a love for AI, and a competitive streak? Bring your skills to McHugh Hall on Mar. 9 for a full day of hacking and learning. Get ready for tech talks on the latest in AI and computer vision, real-world challenges to test your creativity, prizes, swag, and bragging rights for top projects. Whether you're just starting out or ready to flex some serious ML muscle, there’s a track for you. ?? Sign up now: https://lnkd.in/gNvhjiGj #AI #VisualAI #ComputerVision #MachineLearning #Hackathon #AICommunity #DataScience #Connecticut

?? Dive into the depths of underwater object tracking with the groundbreaking WebUOT-1M dataset! The WebUOT-1M dataset is the largest public underwater object tracking benchmark dataset to date. With 1.1 million annotated frames across 1,500 video clips and 408 target categories, this dataset is making waves in visual AI. Want to see it in action? Check out this tutorial and learn how to: ?? Explore WebUOT-1M in FiftyOne ?? Compute and visualize video and text embeddings ?? Apply SAM2 ?? https://lnkd.in/gwcueF_S #VisualAI #AI #MachineLearning #ML #DataScience #DataCentricAI #MarineTech

What’s really powerful about embeddings is that they help solve one of the most critical challenges in working with large-scale datasets for autonomous systems. ?? Finding the rare, unseen, and outlier scenarios that could make or break your model. When building a self-driving dataset, embeddings let you spot gaps, surface edge cases, and prioritize the data that actually moves the needle. With the right approach, you can train smarter, not harder—ensuring your model is ready for the unpredictable realities of the road. ?? Daniel G. breaks down the power of embeddings for data curation here: https://lnkd.in/ecfbZbNq #ComputerVision #VisualAI #AI #MachineLearning #ML #SelfDriving #AutonomousVehicles

What if you could make your annotation process smarter and more efficient? Enter Active Learning. Intelligently select your most crucial samples for labeling, saving time and money while enhancing the overall quality of your models. And we've got a plugin for that! With the Active Learning FiftyOne plugin, you can: ?? Label only what matters most ? Teach your active learner as you go ?? Sync with your annotation service, like CVAT & more Try it now: https://lnkd.in/gJUbzfzK Happy learning! #ComputerVision #VisualAI #AI #AITools #MachineLearning #ML

?? What if memes are actually the perfect way to test vision AI models? Don’t miss the new article by Harpreet Sahota ?? that makes a compelling case: Memes might be the ultimate benchmark for testing VLMs. Think about it - memes are a perfect storm of: ? Visual and text understanding ? Cultural context ? Humor understanding ? Various formats and styles The best part? He put this theory to the test, comparing how different VLMs handle meme-related tasks. Want to see how AI models try to understand memes? Check out the full analysis here: https://lnkd.in/gwdrf_yh #MachineLearning #ML #ArtificialIntelligence #AI #ComputerVision #DataScience #VLM

Synthetic data can be a powerful tool for balancing datasets and mitigating bias. But here's the catch. ?? Using it without analyzing how it actually aligns with real data could distort your model instead of improving it. Take a look at this visualization. These are CLIP embeddings of real (left) vs. synthetic (right) images. At first glance, they look similar. But dive deeper, and you’ll see something surprising: synthetic images don’t always cluster where we expect them to. That misalignment? It can introduce new biases rather than fixing old ones. If you’re working with synthetic data, read on to learn: ?? How embedding visualizations reveal hidden biases in real and synthetic datasets ?? Why some synthetic data samples help while others hurt model performance ?? How to compare and refine synthetic data before it skews your results The insights might surprise you! ?? https://lnkd.in/gh3kpjd5 #AI #VisualAI #ComputerVision #DataCuration #DataCentricAI #MachineLearning #ML #DataScience

We're headed to Boston! And we’re bringing some mind-bending AI topics with us. ?? If you’re curious about soft labels & dataset distillation, lessons from biomedical data, the surprising limits of large-scale pretraining, and more, then you need to be at the Boston AI, ML & Computer Vision Meetup on Feb. 28. Speakers include: ?? Sunny Qin, Harvard University ?? Eric Ma, Moderna ?? Jaedong Hwang, Massachusetts Institute of Technology ?? Harpreet Sahota ?? Sahota, Voxel51 ?? Microsoft Research Lab – New England ?? Feb 28 | 5-8 PM ?? Grab your spot here: https://lnkd.in/gxWKV2BK #AI #VisualAI #AICommunity #ComputerVision #ML #MachineLearning #Boston

Tired of manual data wrangling and cumbersome model evaluation tasks? There’s a better way. ?? Join Harpreet Sahota ?? on Feb. 26 for a hands-on workshop where you'll learn how to: ?? Automate data curation and model evaluation tasks ? Integrate new models, datasets, and MLOps tools—fast ??? Streamline your workflow with less code and more clicks ?? Customize FiftyOne to work the way you want ?? Get the Zoom meeting link here: https://lnkd.in/eDwtv6sw #ComputerVision #ModelEvaluation #VisualAI #MachineLearning #DataCuration #DataScience

Imagine adding classifications, detections, or segmentations to your image datasets—even for scenarios that were not seen during training. That's the power of the FiftyOne Zero-Shot Prediction Plugin. ?? With this plugin, you can: ? Auto-label datasets in seconds ? Skip the manual annotation grind ? Compare results with ground truth—instantly Check out Daniel G.'s tutorial to see it in action, then go try it yourself here: https://lnkd.in/gbghdcX2 #ZeroShotPrediction #ComputerVision #VisualAI #AI #MachineLearning #OpenSource