Model types

There are different types of models for different kind of tasks.

For this project about analyzing a stream of images of the street, what we expect is to use the different types for these purposes:

• semantic segmentation for lane driving, i.e. to figure where the vehicle can ride
• object detection to detect pedestrians, or instance segmentation could be another option, making panoptic segmentation (roughly the addition of semantic and instance segmentation) probably the best option
• image classification is less appropriate, but could be used for global analysis of the road situation or to confirm another method

See Mapbox Vision SDK for examples

See Meta's dataset for object segmentation