IMotionModel

A motion model class receives informations (e.g. frames, estimated flow and depth) and produce an initial guess to the pose of incoming frame under global coordinate.

Interface

Module/MotionModel.py

class IMotionModel(ABC, ConfigTestableSubclass):
    def __init__(self, config: SimpleNamespace):
        self.config : SimpleNamespace = config
    
    @abstractmethod
    def predict(self, frame: SourceDataFrame, flow: torch.Tensor | None, depth: torch.Tensor | None) -> pp.LieTensor:
        ...
    
    @abstractmethod
    def update(self, pose: pp.LieTensor) -> None:
        ...

Methods to Implement

• IMotionModel.predict(...) -> pp.LieTensor

  Estimate the pose of next frame given current frame, estimated depth and flow.

  Returns pose, a (7,) shaped pypose.LieTensor (SE3 ltype) under world coordinate predicted pose of next frame.

• IMotionModel.update(self, pose: pp.LieTensor) -> None

  Receive a feedback (optimized pose) and may (or may not) use this method to refine next prediction.

Implementations

• class GTMotionwithNoise(IMotionModel)

  Apply GT motion with noise (can be disabled by setting noise_std to 0.0 in config) on previous optimized pose to predict next pose.

• ⭐ class TartanMotionNet(IMotionModel)

  Apply motion estimated by MotionNet adapted from TartanVO on previously optimized pose to predict next pose.

• class StaticMotionModel(IMotionModel)

  Assumes the camera is static and simply record and returns the pose of previous frame.

• class ReadPoseFile(IMotionModel)

  Use an external file of Nx7 SE3 poses as motion model output poses.

  info

  Specifically, the module will not output these poses directly but calculate the motion and apply motion on modified poses (potentially by optimizer) iteratively.