The competition with smart Internet + automatic driving, which is at its core, has gone through the growing sound of a software-specific car everywhere, from traditional car companies to new car manufacturing forces.

The core components of cars have gradually evolved into “car brains” of engine, gearbox and chassis chips, software and data. A new competition has been announced. However, a number of new issues are emerging, such as security, laws and regulations.

Tesla’s lead in the field of automatic driving comes mainly from its large data. The vast amount of data and cornering cases generated daily by millions of Tesla cars around the world continuously improve Tesla’s automatic driving function and performance and increase intelligence between Tesla and other automotive companies.

Every other day, Tesla’s intelligence is stronger. When it is strong enough to some extent, there will be a qualitative difference.

The answer is ecology.

Some people think that this year is the first year of high-level automatic driving because we can indeed see great progress in high-level automatic vehicles.

In China before the Shanghai Motor Show

Didi showed urban driverlessness for 5 hours in a row;

Advanced automatic driving test Huawei and Jihu in complex working conditions;

1. Why do business AI projects fail?

2. How will artificial intelligence trigger the next wave of healthcare innovation?

3. Machine learning using a regression model

4. Most popular computing platforms in 2021, other than Kaggle

The first display of the Weiman L4;

Xpeng’s autopilot drives nearly 3,000 kilometers from Guangzhou to Beijing, and the total number of takeovers is about 20, which means about 0.7 to 0.8 times per 100 kilometers.

All of these reflect the overall development of China’s intelligent vehicle automatic driving technology.

However, it is difficult to say whether this year is the first year of high-level automatic driving.

The fact that driving 1000 km is OK does not mean driving a million km, not to mention the whole life cycle. There are difficulties in the safety and reliability of the driver’s operation and it takes a long time to deal with corner cases. The real fall of a high-level automatic drive still requires time and patience.

The first decline in unmanned driving will occur in some special scenarios, such as unmanned trucks in ports and mines, cleaning and logistics equipment in parks, and Robotaxi in certain areas. It will be a long time before the large-scale appearance of unmanned passenger cars.

It is also necessary to remind all car companies to pay attention to propaganda. At present, for example, the Chinese government allows the mass production and application of intelligent driving aid technology, which is limited to L3 or below. Therefore, all car companies should be careful in propaganda and not across borders: don’t let users misunderstand that they are guaranteed to give the right to a car.

Last year at CVPR 2019, Andrej Karpathy, Senior Director of Artificial Intelligence at Tesla, answered the question below:

how to estimate the amount of marked data needed to train and validate self-propelled cars in a given situation? “

378 hours of information.

The more accurate the notation, the better the performance of the algorithm.

TavuSilta is human and ML powered data entry tool platform. We provide scalable, high-quality data for the ML / AI industry with a flexible workflow.

Quality and accuracy

  • ML-assisted capacity can help reduce human error automatically advance markings
  • real-time quality control and quality control integrated into the labeling workflow when a consensus mechanism is put in place to ensure accuracy.
  • Understanding – Assign the same task to multiple employees, and the correct answer is the one that comes back from the majority output.
  • All the results of the work has been fully security checked and inspected machine and manpower.

In this way, ByteBridge can confirm the acceptance and accuracy of the data over 98%.

  • Preview progress: the customer can monitor the progress of the entries in real time on the dashboard
  • Preview the result: the customer can get the results in real time on the dashboard
  • Real-time outputs: the customer can get through real-time outputs API. We support JSON, XML, CSV, etc., and we can provide editable data type to meet your needs.

3D Point Cloud Markup Service

ByteBridge has developed 3D Point Cloud markings, a quality control tool, and label functions that can complement high-quality and accurate 3D point cloud markings for 2D-3D fusion or 3D images provided by different manufacturers and devices, and provide a single station management service, markings, quality control and quality control.

More information: ByteBridge launches the world’s first mobile 3D Point cloud service tag service

3D point cloud annotation types:

  • Sensor Fusion Cuboids: 12 categories include car, truck, heavy vehicle, two-wheeled vehicle, pedestrian, etc.
  • Sensor fusion segmentation: barrier classification, different types of band specifications
  • Monitoring of sensor fusion cuboids

① Tracking the same object with the same ID, marking the initial state;

② Dot clouds or time-aligned images can be output, only dot cloud outputs.

The collaboration of human and labor and artificial intelligence algorithms ensures a 50% lower price compared to the normal market.

Because the quality of the marked data sets determines the success of the self-driving industry, collaboration reliable partner can help developers overcome data labeling challenges.

If you need data entry and collection services, check out bytebridge.io, clear pricing is available.

Feel free to contact us: support@bytebridge.io

LEAVE A REPLY

Please enter your comment!
Please enter your name here