Modern mobility systems rely on massive, high-quality multimodal datasets — yet real-world data is messy. Misaligned sensors, inconsistent metadata, and uneven scenario coverage can slow development and lead to costly model failures. The Physical AI Workbench, built in collaboration between Voxel51 and NVIDIA, provides an automated and scalable pipeline for auditing, reconstructing, and enriching autonomous driving datasets.

In this talk, we’ll show how FiftyOne serves as the central interface for inspecting and validating sensor alignment, scene structure, and scenario diversity, while NVIDIA Neural Reconstruction (NuRec) enables physics-aware reconstruction directly from real-world captures. We’ll highlight how these capabilities support automated dataset quality checks, reduce manual review overhead, and streamline the creation of richer datasets for model training and evaluation.

Attendees will gain insight into how Physical AI workflows help mobility teams scale, improve dataset reliability, and accelerate iteration from data capture to model deployment — without rewriting their infrastructure.

Modern mobility systems rely on massive, high-quality multimodal datasets — yet real-world data is messy. Misaligned sensors, inconsistent metadata, and uneven scenario coverage can slow development and lead to costly model failures. The Physical AI Workbench, built in collaboration between Voxel51 and NVIDIA, provides an automated and scalable pipeline for auditing, reconstructing, and enriching autonomous driving datasets.\n\nIn this talk, we’ll show how FiftyOne serves as the central interface for inspecting and validating sensor alignment, scene structure, and scenario diversity, while NVIDIA Neural Reconstruction (NuRec) enables physics-aware reconstruction directly from real-world captures. We’ll highlight how these capabilities support automated dataset quality checks, reduce manual review overhead, and streamline the creation of richer datasets for model training and evaluation.\n\nAttendees will gain insight into how Physical AI workflows help mobility teams scale, improve dataset reliability, and accelerate iteration from data capture to model deployment — without rewriting their infrastructure.

Modern mobility systems rely on massive, high-quality multimodal datasets — yet real-world data is messy. Misaligned sensors, inconsistent metadata, and uneven scenario coverage can slow development and lead to costly model failures. The Physical AI Workbench, built in collaboration between Voxel51 and NVIDIA, provides an automated and scalable pipeline for auditing, reconstructing, and enriching autonomous driving datasets. In this talk, we’ll show how FiftyOne serves as the central interface for inspecting and validating sensor alignment, scene structure, and scenario diversity, while NVIDIA Neural Reconstruction (NuRec) enables physics-aware reconstruction directly from real-world captures. We’ll highlight how these capabilities support automated dataset quality checks, reduce manual review overhead, and streamline the creation of richer datasets for model training and evaluation. Attendees will gain insight into how Physical AI workflows help mobility teams scale, improve dataset reliability, and accelerate iteration from data capture to model deployment — without rewriting their infrastructure.

Modern mobility systems rely on massive, high-quality multimodal datasets — yet real-world data is messy. Misaligned sensors, inconsistent metadata, and uneven scenario coverage can slow development and lead to costly model failures. The Physical AI Workbench, built in collaboration between Voxel51 and NVIDIA, provides an automated and scalable pipeline for auditing, reconstructing, and enriching autonomous driving datasets. In this talk, we’ll show how FiftyOne serves as the central interface for inspecting and validating sensor alignment, scene structure, and scenario diversity, while NVIDIA Neural Reconstruction (NuRec) enables physics-aware reconstruction directly from real-world captures. We’ll highlight how these capabilities support automated dataset quality checks, reduce manual review overhead, and streamline the creation of richer datasets for model training and evaluation. Attendees will gain insight into how Physical AI workflows help mobility teams scale, improve dataset reliability, and accelerate iteration from data capture to model deployment — without rewriting their infrastructure.

Modern mobility systems rely on massive, high-quality multimodal datasets — yet real-world data is messy. Misaligned sensors, inconsistent metadata, and uneven scenario coverage can slow development and lead to costly model failures. The Physical AI Workbench, built in collaboration between Voxel51 and NVIDIA, provides an automated and scalable pipeline for auditing, reconstructing, and enriching autonomous driving datasets. In this talk, we’ll show how FiftyOne serves as the central interface for inspecting and validating sensor alignment, scene structure, and scenario diversity, while NVIDIA Neural Reconstruction (NuRec) enables physics-aware reconstruction directly from real-world captures. We’ll highlight how these capabilities support automated dataset quality checks, reduce manual review overhead, and streamline the creation of richer datasets for model training and evaluation. Attendees will gain insight into how Physical AI workflows help mobility teams scale, improve dataset reliability, and accelerate iteration from data capture to model deployment — without rewriting their infrastructure.

We present a multimodal AI pipeline to streamline patient selection and quality assessment for radiology AI development. Our system evaluates patient clinical histories, imaging protocols, and data quality, embedding results into imaging metadata. Using FiftyOne researchers can rapidly filter and assemble high-quality cohorts in minutes instead of weeks, freeing radiologists for clinical work and accelerating AI tool development.

We present a multimodal AI pipeline to streamline patient selection and quality assessment for radiology AI development. Our system evaluates patient clinical histories, imaging protocols, and data quality, embedding results into imaging metadata. Using FiftyOne researchers can rapidly filter and assemble high-quality cohorts in minutes instead of weeks, freeing radiologists for clinical work and accelerating AI tool development.

We present a multimodal AI pipeline to streamline patient selection and quality assessment for radiology AI development. Our system evaluates patient clinical histories, imaging protocols, and data quality, embedding results into imaging metadata. Using FiftyOne researchers can rapidly filter and assemble high-quality cohorts in minutes instead of weeks, freeing radiologists for clinical work and accelerating AI tool development.

We present a multimodal AI pipeline to streamline patient selection and quality assessment for radiology AI development. Our system evaluates patient clinical histories, imaging protocols, and data quality, embedding results into imaging metadata. Using FiftyOne researchers can rapidly filter and assemble high-quality cohorts in minutes instead of weeks, freeing radiologists for clinical work and accelerating AI tool development.

We present a multimodal AI pipeline to streamline patient selection and quality assessment for radiology AI development. Our system evaluates patient clinical histories, imaging protocols, and data quality, embedding results into imaging metadata. Using FiftyOne researchers can rapidly filter and assemble high-quality cohorts in minutes instead of weeks, freeing radiologists for clinical work and accelerating AI tool development.

We present a multimodal AI pipeline to streamline patient selection and quality assessment for radiology AI development. Our system evaluates patient clinical histories, imaging protocols, and data quality, embedding results into imaging metadata. Using FiftyOne researchers can rapidly filter and assemble high-quality cohorts in minutes instead of weeks, freeing radiologists for clinical work and accelerating AI tool development.

Day 2 focuses on visual dataset curation with FiftyOne and iterative improvement of image classification models.

Focus on building and training neural networks with PyTorch.

Focus on visual dataset curation with FiftyOne and iterative improvement of image classification models.

Day 2 focuses on visual dataset curation with FiftyOne and iterative improvement of image classification models.

DeepSeek’s Janus-Pro is an advanced multimodal model designed for both multimodal understanding and visual generation, with a particular emphasis on improvements in understanding tasks. In this talk, we’ll explore Janus-Pro’s Visual Question Answer (VQA) capabilities using FiftyOne’s Janus-Pro VQA Plugin. The plugin provides a seamless interface to Janus Pro’s visual question understanding capabilities within FiftyOne, offering: Vision-language tasks; Hardware acceleration (CUDA/MPS) when available; Dynamic version selection from HuggingFace; Full integration with FiftyOne’s Dataset and UI. Can’t wait to see it for yourself? Check out the FiftyOne Quickstart with Janus-Pro.

Exploration of Janus-Pro’s VQA capabilities using FiftyOne’s Janus-Pro VQA Plugin, with a focus on vision-language tasks, hardware acceleration (CUDA/MPS), dynamic version selection from HuggingFace, and full integration with FiftyOne’s Dataset and UI.

In this talk, we’ll explore Janus-Pro’s Visual Question Answer (VQA) capabilities using FiftyOne’s Janus-Pro VQA Plugin. The plugin provides a seamless interface to Janus Pro’s visual question understanding capabilities within FiftyOne, offering: Vision-language tasks; Hardware acceleration (CUDA/MPS) when available; Dynamic version selection from HuggingFace; Full integration with FiftyOne’s Dataset and UI.

Overview of maximizing the potential of the FiftyOne open-source SDK and App to efficiently store and annotate training data for Generative AI workflows, including cloud-based storage/hosting architecture, training/apply models for semi-automatic data annotation, and CVAT integration for pixel-perfect side-by-side evaluation.

In this talk we’ll explore how we maximize the potential of the FiftyOne open source SDK and App to efficiently store and annotate training data critical to Finegrain’s Generative AI workflows. We will provide an overview of our cloud-based storage and hosting architecture, showcase how we leverage FiftyOne for training and applying models for semi-automatic data annotation, and demonstrate how we extend the CVAT integration to enable pixel-perfect side-by-side evaluation of our Generative AI models.