# Map Optimization

For map customization and optimizations you can tweak these parameters in the script and reconstruct the map again with custom parameters. These parameters given below can be changed.

def StartMapConstruction(url: str, token: str, mapName: str) -> str:
    complete_url = url + "/construct"

    data = {
        "token": token,
        "name": mapName,
        "preservePoses": False,
        # "featureCount": 1024, #default: 1024
        # "featureCountMax": 8192, #default: 8192,
        # "featureFilter": 0, # default: 0
        # "trackLengthMin": 2, #default: 2
        # "triangulationDistanceMax": 512, # default: 512
        # "dense": 1, # default: 1
    }

You may pass several parameters to ‘/construct’ endpoint of REST API for tweaking the map construction

(default=1024): Maximum num of features/landmarks to keep per image after map construction.

(default=8192): Maximum num of features extracted from image. By increasing these value, more feature will be extracted, but it will also generate bigger map.

(default=0) Possible values 0 or 1, in scenes where there’s lots of unique details like grass, bushes, leaves, gravel etc we’d pick a lot of noisy details (high frequency features) as our top picks. These features were very hard to localize against. By using this parameter in map construction, it sorts the detected features based on size favoring large features (low frequency features). This seems to improve map construction and localization rate in high frequency environments.

(default=2) This value represents the minimum number of images from which a feature point must originate to be kept in the map. The larger the number, the higher the reliability required for feature points, resulting in fewer points being retained and a smaller map. We usually start with a default value of 2 and gradually increase it (typically between 2 and 5), observing the success rate of positioning until it noticeably drops.

This value represents the maximum distance to the target object for positioning, measured in meters. A value of 512 means that the system supports constructing a point cloud for an object up to 512 meters away. It's important to note that in triangulation, if your target object is far away, your baseline (the distance moved laterally) should be correspondingly increased, typically to 5%-10% of the distance. Otherwise, the point clouds for these distant objects will be unreliable. It is also important to note that sometimes constructing point clouds for distant objects is not a good idea because the accuracy of positioning decreases with distance. Therefore, you should adjust this parameter based on the actual scenario.