vis4d.op

Compositional operators used for implementing models.

This is where most of the library APIs are implemented. All the operators are functors. They are native PyTorch modules and only have a forward member for function invocations. We follow the principle of functional programming. The operators don’t keep internal states besides the operator weights. The operator computation and call has no side effects.

Modules

vis4d.op.base	Base model module.
vis4d.op.box	Operations on 2D bounding boxes.
vis4d.op.detect	Detector module.
vis4d.op.detect3d	3D detector module.
vis4d.op.fpp	Vis4D modules for feature pyramid processing.
vis4d.op.geometry	Init geometry module.
vis4d.op.layer	Init layers module.
vis4d.op.loss	This module contains commonly used loss functions.
vis4d.op.mask	Operations on 2D segmentation masks.
vis4d.op.motion	Motion operations.
vis4d.op.seg	Segmentor module.
vis4d.op.track	Tracking models module.
vis4d.op.track3d	3D tracking models module.
vis4d.op.typing	Vis4D op typing.
vis4d.op.util	Utilities for op.