At present, the mainstream algorithm models of auto-driving are mainly supervised deep learning.
Supervised learning is the machine learning task of learning a function that maps an input to an output based on example input-output pairs. It infers a function from labeled training data consisting of a set of training examples.
As it is based on known variables and dependent variables to derive function relations. A large amount of structured annotated data is needed to train and tune the model.
Common annotation types in auto-driving usually include 2D bounding box, 3D cube bounding box, polygon, image segmentation, video annotation, 3D laser point cloud annotation, etc., among which image semantic segmentation is widely used.
Conceptually, image semantic segmentation is an important annotation type in computer vision. It includes image classification, target detection, image segmentation, and mainly segmentation at the pixel level.
2. Deep Learning in Self-Driving Cars
3. Generalization Technique for ML models
4. Why You Should Ditch Your In-House Training Data Tools (And Avoid Building Your Own)
The result of semantic segmentation is to transform the image into several color blocks, and each color block represents one part of the image.
These annotated segmentation images can be used to train the algorithm of automatic driving. During the driving process of the self-driving car, the images detected by the on-board camera or radar are input into the neural network system, and the well-trained algorithm model can automatically segment and classify the images, so as to avoid obstacles such as pedestrians and vehicles on the road.
Currently, in the field of semantic segmentation of auto-driving images, the commonly used annotation objects mainly are into the following categories
01. Accessible Road areas
Note: The driving zone usually refers to the area where cars can drive. When there are curbs, the driving area is the part between curbs. When curbs are not present, the asphalt or other covered area shall prevail.
02. Handrails/barrier/stone pier
Note: located on both sides of the road, mainly including railings, barriers, stone pier, and other isolators, but does not include temporary roadblocks set up because of road construction.
03. Rod (the vertical part)
Note: only the vertical part of the rod should be marked, other parts of the rod should not be marked. They usually include street signposts, street light posts, traffic light posts, etc.
04. Roadside buildings
Notes: buildings on both sides of the road, including high-rise buildings and low-level buildings, and other man-made building objects.
ByteBridge, a human-powered data labeling tooling platform with real-time workflow management, providing high-quality data with efficiency.
Having an R&D team with rich development resources, ByteBridge can provide personalized annotation tools and services according to customer requirements.
Regarding flexibility, clients can set labeling rules, iterate data features, attributes, and task flows, scale up or down, make changes.
Moreover, clients can monitor the labeling progress and get the results in real-time on the visualized dashboard.
For more information, please have a look at bytebridge.io, the clear pricing is available.