preprocess/humanparsing/mhp_extension/detectron2/docs/tutorials/write-models.md

Write Models

If you are trying to do something completely new, you may wish to implement a model entirely from scratch within detectron2. However, in many situations you may be interested in modifying or extending some components of an existing model. Therefore, we also provide a registration mechanism that lets you override the behavior of certain internal components of standard models.

For example, to add a new backbone, import this code in your code:

from detectron2.modeling import BACKBONE_REGISTRY, Backbone, ShapeSpec

@BACKBONE_REGISTRY.register()
class ToyBackBone(Backbone):
  def __init__(self, cfg, input_shape):
    # create your own backbone
    self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=16, padding=3)

  def forward(self, image):
    return {"conv1": self.conv1(image)}

  def output_shape(self):
    return {"conv1": ShapeSpec(channels=64, stride=16)}

Then, you can use cfg.MODEL.BACKBONE.NAME = 'ToyBackBone' in your config object. build_model(cfg) will then call your ToyBackBone instead.

As another example, to add new abilities to the ROI heads in the Generalized R-CNN meta-architecture, you can implement a new ROIHeads subclass and put it in the ROI_HEADS_REGISTRY. See densepose in detectron2 and meshrcnn for examples that implement new ROIHeads to perform new tasks. And projects/ contains more examples that implement different architectures.

A complete list of registries can be found in API documentation. You can register components in these registries to customize different parts of a model, or the entire model.