Parallel Domain
Parallel Domain
  • 24
  • 9 587
Reconstructing a known Tesla FSD accident to understand perception failures
This Tesla FSD accident happened on June 13, 2024 in Fullerton, CA.
Using Parallel Domain we can reconstruct the scenario and simulate thousands of variations to identify which variables contributed most to the perception failure.
As autonomous systems grow in real-world deployment, simulation is no longer a “nice to have”-it’s essential for uncovering hidden performance gaps and testing against thousands of mission critical scenarios daily to ensure safe, reliable performance.
Curious how your perception model performs under pressure? Test it out with a free trial of Parallel Domain-and see how your system stacks up. paralleldomain.com/automotive
มุมมอง: 144

วีดีโอ

This is a simulation. Closing the real-to-sim gap with simulation capable scene reconstructions
มุมมอง 169หลายเดือนก่อน
All of this video is a simulation. At Parallel Domain, we are taking simulation to environments rarely digitized in such high-fidelity! 🌍 Introducing our brand new Farm and Warehouse PD Replica locations, complementing our extensive collection of road, parking, and aerial environments. What are PD Replica locations? PD Replica locations are simulation-capable digital twins generated from photos...
ECCV Oral Session - Generative Camera Dolly:Extreme Monocular Dynamic Novel View Synthesis
มุมมอง 762 หลายเดือนก่อน
We are proud to have played a small role in this groundbreaking research! Through our partnership with Toyota Research Institute we provided 2.4 Tb of synthetic data, comprised of 1500 scenes covering diverse environments, traffic patterns, vehicles, pedestrians, and weather conditions. Each scene contained synchronized videos from 19 camera viewpoints, with RGB, depth, optical flow, surface no...
Emergency Vehicle simulations created in a Parallel Domain
มุมมอง 633 หลายเดือนก่อน
Pulling over cars daily-in a Parallel Domain! 🌐 Ensuring that no matter where in the world machine perception models are operating, they recognize the vehicles and lights of emergency services. Excited to share our fun video showcasing a diverse array of emergency vehicles 🚓 🚑 🚒 from around the world in various scenarios and environments! Those environments can be both procedurally generated or...
This Flight Never Happened - Simulation in a PD Replica
มุมมอง 403 หลายเดือนก่อน
This flight never happened. Utilizing PD Replica one can use a real-world scan as the environment to conduct high fidelity simulations to output RGB, LiDAR, and Radar data for training and testing perception systems at scale.
Automotive Simulation in a Digital Twin - PD Replica driving compilation
มุมมอง 1023 หลายเดือนก่อน
In this video all of the locations you see are real, however the driving never happened. All of these are output renderings from the Parallel Domain platform contain RGB with full annotations including bounding boxes, depth, points, instance and semantic segmentation. Data like this gets machine learning teams one step closer to testing and training in the real-world. Replica fits into Parallel...
Real-world AI testing in a pd replica, simulation-ready digital twins + synthetic data generation
มุมมอง 9446 หลายเดือนก่อน
pd replica is software to create simulation-ready, pixel-perfect 3d digital copies of real world locations from photos! Existing digital twin technology does not allow both synthetic scenario simulation and this level of fidelity. It’s a giant step forward towards the matrix. Testing in the real world is extremely difficult to do, expensive, and risky. This technology enables teams to mimic the...
Data Lab: craft your optimal synthetic dataset, supercharged by generative AI.
มุมมอง 567ปีที่แล้ว
Data Lab: The ultimate control hub for crafting your optimal synthetic dataset, supercharged by generative AI paralleldomain.com/products/data-lab
Simulation and Domain Adaptation at Cruise - Ashish Shrivastava CVPR SDAS 2023
มุมมอง 542ปีที่แล้ว
Simulation plays a crucial role in the development of autonomous vehicles at Cruise as they expand to new platforms and cities. By utilizing high-fidelity virtual environments, they can effectively simulate a wide range of scenarios, including challenging edge cases, adverse weather conditions, and rare events that are seldom encountered in real-world situations. A key highlight of the talk is ...
Data Sparsity Challenges for Pedestrian Understanding with Synthetic Data -Junhua Mao CVPR SDAS 2023
มุมมอง 191ปีที่แล้ว
Junhua Mao from Waymo presents a few methods to tackle data sparsity, including: - Data augmentation and synthetic data - Weak supervision from relatively easy-to-get labels - Self-supervision and joint training
Principle-centric Machine Learning for Embodied Foundation Models - Dr. Adrien Gaidon CVPR SDAS 2023
มุมมอง 185ปีที่แล้ว
Can Machine Learning's success on the web translate to intelligent machines physically interacting with the real world, especially for safety-critical systems like cars and robots? In this talk, Adrien describes their unique approach to the problem of Embodied Intelligence: Principle-centric ML. It consists in going beyond data by complementing it with guiding principles to increase safety, gen...
Simulation For Semantic Self-Labeling and Self-Driving - Dr. Antonio M. López CVPR SDAS 2023
มุมมอง 197ปีที่แล้ว
In this workshop, Dr. López discusses how: - The creation of a photo-realistic synthetic dataset which, when combined with other synthetic datasets and semi-supervised learning techniques, can be used to get state-of-the-art results in onboard semantic segmentation - Using a digital twin and the CARLA simulator allowed autonomous driving on narrow mountain roads
Seasoning the World with Simulation: Self-Driving Improvement with Synthetic Data - CVPR SDAS 2023
มุมมอง 654ปีที่แล้ว
A deep dive from Charles Henden on how the sim team at Tesla has fully embraced a procedural methodology built on real-world maps and how they built tooling from the ground up to reach our goal of truly becoming an infinite data machine. With the ability to fuse real-world locations with driving scenarios mined from fleet data, they produce synthetic datasets that have parameters permuted speci...
Improving Network Architectures & Training for Semantic Segmentation - Lukas Hoyer CVPR SDAS 2023
มุมมอง 309ปีที่แล้ว
As acquiring pixel-wise annotations of real-world images for semantic segmentation is costly, a model can instead be trained with more accessible synthetic data and adapted to real images without requiring their annotations. This process is studied in unsupervised domain adaptation. Most research in these areas focused on the design of adaptation strategies to overcome the problem of the domain...
On the Importance of Label Domain Gaps - Phillip Thomas CVPR SDAS 2023
มุมมอง 531ปีที่แล้ว
Why is it so hard to make synthetic data work in conjunction with real data? Even though there is a lot of research on the topic of Domain Adaptation, there is not yet a common definition of what the term domain gap entails. In this talk, Phillip Thomas from Parallel Domain presents a Domain Gap Ontology and shows how gaps in the Label Domain decide if a model can benefit from synthetic data.
Unlocking AI Perception Gains with Iterative Synthetic Data Generation
มุมมอง 446ปีที่แล้ว
Unlocking AI Perception Gains with Iterative Synthetic Data Generation
Parking: Synthetic Data Showcase
มุมมอง 377ปีที่แล้ว
Parking: Synthetic Data Showcase
Trailers: Synthetic Data Showcase
มุมมอง 319ปีที่แล้ว
Trailers: Synthetic Data Showcase
Road Debris: Synthetic Data Showcase
มุมมอง 460ปีที่แล้ว
Road Debris: Synthetic Data Showcase
Road Signs: Synthetic Data Showcase
มุมมอง 432ปีที่แล้ว
Road Signs: Synthetic Data Showcase
Parallel Domain Web Visualizer Tutorial
มุมมอง 250ปีที่แล้ว
Parallel Domain Web Visualizer Tutorial
Parallel Domain | Accelerate Perception Development with Synthetic Data
มุมมอง 787ปีที่แล้ว
Parallel Domain | Accelerate Perception Development with Synthetic Data
Synthetic Data Workshop: Open Dataset Release for Improving Cyclist Detection
มุมมอง 6573 ปีที่แล้ว
Synthetic Data Workshop: Open Dataset Release for Improving Cyclist Detection
How TRI Trains Better Computer Vision Models with PD Synthetic Data
มุมมอง 1.2K3 ปีที่แล้ว
How TRI Trains Better Computer Vision Models with PD Synthetic Data

ความคิดเห็น

  • @sandippatil8866
    @sandippatil8866 ปีที่แล้ว

    Great work

  • @randalcurtis6667
    @randalcurtis6667 ปีที่แล้ว

    Promo SM 👊

  • @MsSabhishek
    @MsSabhishek ปีที่แล้ว

    Great Ashish 👍

  • @MsSabhishek
    @MsSabhishek ปีที่แล้ว

    Good 👍

  • @paralleldomain3649
    @paralleldomain3649 ปีที่แล้ว

    Timestamps: 00:00: Introducing Omar and Phillip 01:12: How to Get Value Out of Synthetic Data 02:06: Agenda 02:54: Problems with Current ML Development Practices 04:23: Edge Case Detection using Synthetic Data 05:11: Parallel Domain’s Image Data (parking, trailers, debris, road signs, etc.) 05:45: Parallel Domain’s Sensors and Annotation types (labels) 06:19: Performance Improvements using Synthetic Data 07:45: Iterative Synthetic Dataset Design 09:15: Domain Gap Ontology 25:29 Parallel Domain SDK 28:46 Demo - using Synthetic Data for 2D Object Detection 32:11 Demo - Performance Iteration with Synthetic Data 35:11 Dataset Design - Optical Flow Example 39:54 Dataset Design - Semantic Segmentation UDA Example 43:43 Demo - Reducing Synthetic Data Iteration Times using Statistics 47:41 Why Parallel Domain 48:44 Parallel Domain at CVPR 51:24 Questions and Answers