mask_rcnn_resnet_101_fpn¶

lucid.models.mask_rcnn_resnet_101_fpn(num_classes: int = 21, backbone_num_classes: int = 1000, **kwargs) → MaskRCNN¶

The mask_rcnn_resnet_101_fpn function builds a Mask R-CNN instance segmentation model with a deeper ResNet-101 backbone and FPN features.

Total Parameters: 65,126,611

Function Signature¶

@register_model
def mask_rcnn_resnet_101_fpn(
    num_classes: int = 21,
    backbone_num_classes: int = 1000,
    **kwargs
) -> MaskRCNN

Parameters¶

num_classes (int, optional): Number of segmentation/detection classes. Default is 21.
backbone_num_classes (int, optional): Number of classes used to initialize the internal ResNet-101 backbone.
kwargs (dict, optional): Additional keyword arguments passed to MaskRCNNConfig. This factory fixes the ResNet-101 FPN backbone preset and the default feature width.

Returns¶

MaskRCNN: Mask R-CNN model configured with ResNet-101 + FPN.

Example Usage¶

from lucid.models.vision.mask_rcnn import mask_rcnn_resnet_101_fpn
import lucid

model = mask_rcnn_resnet_101_fpn(num_classes=21)
x = lucid.random.randn(1, 3, 640, 640)

cls_logits, bbox_deltas, mask_logits = model(x)
print(cls_logits.shape, bbox_deltas.shape, mask_logits.shape)

out = model.predict(x)
print(out[0]["boxes"].shape, out[0]["masks"].shape)