HyperAI

Will Carrot Run Be Profitable Next Year? Autonomous Driving Opens a New Era of "end-to-end" and High-quality Data Sets Help AI Big Models Be Put Into Cars

a year ago
Information
zhaorui
特色图像

At Baidu Apollo Day 2024 held on May 15, Chen Zhuo, general manager of Baidu's autonomous driving business unit, said that the goal of LuoBoKuaiPao is to achieve a break-even in Wuhan by the end of 2024.It will enter the full profitability period in 2025.Perhaps it was driven by this profit target that Luobo Kuaipao recently started large-scale operations in Wuhan, which also attracted widespread attention.

Putting aside the controversy in the center of public opinion about "driverless taxis stealing jobs", the low starting price and low unit price of Luobo Kuaipao do bring certain competitiveness, and the operating model explored by Baidu also provides a strong reference for the implementation and development of "end-to-end" autonomous driving solutions to a certain extent.

August 2023Tesla's FSD V12 system officially brings "end-to-end" autonomous driving technology to mass-produced models.It also brought it into the public eye. Subsequently, Huawei, Xiaopeng, SenseTime, Baidu and others followed suit, and a new industry technology paradigm is moving from the exploration stage to mature application.

The so-called "end-to-end" solution regards the entire intelligent driving system as an integrated module.After receiving the input data from the sensor, the system directly outputs the driving decision. Its development model has also shifted from rule-driven to data-driven. That is, through training with a large amount of valuable data, AI can autonomously learn human driving patterns until intelligence emerges.

This means that the further development of autonomous driving technology is inseparable from massive high-quality data training. In June 2023, a Tesla software engineer said in a speech at the CVPR conference that "for training the basic model of autonomous driving, we do not require an unlimited amount of data, but we do require diversity based on a certain level."

As the leading search engine in the field of data science in China, HyperAI Super Neural (hyper.ai),We have also paid attention to the high-quality data needs in the field of autonomous driving, and provided you with accelerated downloads of popular open source autonomous driving datasets. Some datasets are summarized below.

Click to view more open source datasets:
https://go.hyper.ai/5Ik29

Autonomous driving dataset

1. ApolloScape autonomous driving dataset

Publishing Agency:Baidu

Release time:2018

Download address:go.hyper.ai/CeKea
ApolloScape is part of the Apollo autonomous driving project, which aims to promote innovation in all areas of autonomous driving, from perception, navigation to control. The dataset is still being updated in terms of data size, label density, and tasks.

2. SODA10M autonomous driving dataset

Publishing Agency:Huawei Noah's Ark Laboratory, Sun Yat-sen University

Release time:2021

Estimated size:5.61 GB

Download address:go.hyper.ai/B3XhR

SODA10M is a semi-/self-supervised 2D benchmark dataset, which mainly contains 10 million diverse unlabeled road scene images and 20,000 images labeled with 6 representative object categories. The images also contain a variety of different road scenes (city, highway, urban and rural roads, parks), weather (sunny, cloudy, rainy, snowy), time periods (daytime, night, early morning/dusk).

3. Talk2Car Autonomous Driving Dataset

Publishing Agency:Catholic University of Leuven

Release time:2020

Estimated size:77.6 MB

Download address:go.hyper.ai/jYVHt
The Talk2Car dataset is an object reference dataset that contains commands written in natural language for self-driving cars, that is, passengers can give commands to self-driving cars by speaking. The Talk2Car dataset is built on the nuScenes dataset and includes a wide set of sensor modalities, namely semantic maps, GPS, LiDAR, radar, and 360° RGB images with 3D bounding box annotations.

4. A2D2 Audi Autonomous Driving Dataset

Publishing Agency:Audi

Release time:2020

Estimated size:2.26 TB

Download address:go.hyper.ai/K1iRu

This dataset is an Audi autonomous driving dataset that contains synchronized images and 3D point clouds, as well as 3D bounding boxes, semantic segmentation, instance segmentation, and data extracted from vehicle buses.

5. Argoverse autonomous driving
Publishing Agency:Argo AI, Carnegie Mellon University, Georgia Institute of Technology

Release time:2019

Estimated size:260.38 GB

Download address:https://go.hyper.ai/CpqTd

The Argoverse dataset is obtained from more than 1,000 hours of street driving and includes two parts: 3D Tracking and Motion Forecasting. The Argoverse 3D tracking dataset contains 3D tracking annotations for 113 scenes. Each clip is 15-30 seconds long and contains a total of 11,319 tracked objects.

6. Lyft Level 5 Autonomous Driving Dataset

Publishing Agency:Lyft

Release time:2019

Estimated size:41.59 GB

Download address:go.hyper.ai/Q3cmA

The Lyft L5 autonomous driving dataset is an L5 autonomous driving dataset provided by Lyft. Currently, only the training set is available for download. The dataset contains high-quality semantic maps and provides detection of the existence and movement of objects. The dataset provides map information such as more than 4,000 roads, 197 crosswalks, 60 stop signs, and 54 parking areas.

7. BLVD Large 5D Semantic Benchmark Dataset

Publishing Agency:Institute of Artificial Intelligence and Robotics, Xi'an Jiaotong University

Release time:2019

Estimated size:43.38 GB

Download address:go.hyper.ai/4ft0g

This dataset is the world's first five-dimensional driving scene understanding dataset, containing 654 high-resolution video clips with a total of 120,000 frames, including 249,129 3D annotated frames, 4,902 independent frames for tracking (with a total length of 214,922 points), 6,004 valid clips for 5D interactive event recognition, and 4,900 frames for 5D intent prediction.

8. DBNet autonomous driving dataset

Publishing Agency:Xiamen University, Shanghai Jiao Tong University

Release time:2018

Estimated size:9.47 GB

Download address:go.hyper.ai/NPXIL

DBNet is a large-scale dataset for driving behavior research. The dataset includes aligned videos, point clouds, GPS, and driver behavior (speed and wheel motion trajectories), capturing 1,000 kilometers of real driving data.

9. JAAD Autonomous Driving Dataset

Publishing Agency:New York University

Release time:2017

Estimated size:2.88 GB

Download address:https://go.hyper.ai/mghSj

The dataset contains 346 video clips, each with a duration of 5-10 seconds and a frame rate of 30, for a total of 82,032 frames. The videos are collected by three on-board cameras and cover typical scenes of daily urban driving in various weather conditions in North America and Eastern Europe.

10. Comma.ai Video Dataset

Publishing Agency:Comma.ai

Release time:2016

Estimated size:44.96 GB

Download address:https://go.hyper.ai/mvkTH

The dataset contains 10 videos recorded at 20Hz, totaling 7.25 hours, which were recorded by a camera mounted on the windshield of an Acura ILX 2016.

The above are the 10 autonomous driving data sets compiled by HyperAI. If you have resources that you want to include on the hyper.ai official website, you are welcome to leave a message or submit an article to tell us!

About HyperAI

HyperAI (hyper.ai) is the leading artificial intelligence and high-performance computing community in China.We are committed to becoming the infrastructure in the field of data science in China and providing rich and high-quality public resources for domestic developers. So far, we have:

* Provide domestic accelerated download nodes for 1200+ public data sets

* Includes 300+ classic and popular online tutorials

* Interpretation of 100+ AI4Science paper cases

* Support 500+ related terms search

* Hosting the first complete Apache TVM Chinese documentation in China

Visit the official website to start your learning journey:

https://go.hyper.ai/8atMX