Image Labeling and Description for Autonomous Driving
Safety is a priority for an autonomous driving engineering team. The vehicle must learn how to avert all risks in any situations. Many autonomous vehicle manufacturers worldwide collect large amounts of video/imagery from the streets. To keep their vehicles and passengers safe, these manufacturers need to turn this raw, unstructured data into high-quality machine learning training data by identifying all potential traffic risk agents.
DataForce set up an image labeling and annotation project with a specifically tailored UI on its proprietary and secure AI data platform. Our annotators analyzed and annotated 18,000 assets that included real-world images captured by cameras on test vehicles. DataForce created a high-quality machine learning training data set by annotating pedestrians, cyclists, incoming traffic, and other critical real-life traffic agents. In addition, we added the following metadata:
• A caption describing the video
• Type of scene that the vehicle is in
• Type of visible agents and expected moving directions • Presence of risk for the vehicle
• Recommended action for the vehicle
The client’s engineering team reviewed DataForce’s work directly on our proprietary annotation platform so that they could share their comments with the annotation team in real time. TransPerfect DataForce completed the image labeling and annotation of 18,000 assets within four weeks and with only five days’ notice.