Skip to content

deeplabcut.pose_estimation_pytorch.models.detectors.fasterRCNN

Classes:

Name Description
FasterRCNN

A FasterRCNN detector.

FasterRCNN

Bases: TorchvisionDetectorAdaptor

A FasterRCNN detector.

Towards Real-Time Object Detection with Region Proposal Networks

Ren, Shaoqing, Kaiming He, Ross Girshick, and Jian Sun. "Faster r-cnn: Towards real-time object detection with region proposal networks." Advances in neural information processing systems 28 (2015).

This class is a wrapper of the torchvision implementation of a FasterRCNN (source: https://github.com/pytorch/vision/blob/main/torchvision/models/detection/faster_rcnn.py).

Some of the available FasterRCNN variants (from fastest to most powerful): - fasterrcnn_mobilenet_v3_large_fpn - fasterrcnn_resnet50_fpn - fasterrcnn_resnet50_fpn_v2

Parameters:

Name Type Description Default

variant

str

The FasterRCNN variant to use (see all options at https://pytorch.org/vision/stable/models.html#object-detection).

 'fasterrcnn_mobilenet_v3_large_fpn'

pretrained

bool

Whether to load model weights pretrained on COCO

 False

box_score_thresh

float

during inference, only return proposals with a classification score greater than box_score_thresh

 0.01
Source code in deeplabcut/pose_estimation_pytorch/models/detectors/fasterRCNN.py 
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
@DETECTORS.register_module
class FasterRCNN(TorchvisionDetectorAdaptor):
    """A FasterRCNN detector.

    Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
        Ren, Shaoqing, Kaiming He, Ross Girshick, and Jian Sun. "Faster r-cnn: Towards
        real-time object detection with region proposal networks." Advances in neural
        information processing systems 28 (2015).

    This class is a wrapper of the torchvision implementation of a FasterRCNN (source:
    https://github.com/pytorch/vision/blob/main/torchvision/models/detection/faster_rcnn.py).

    Some of the available FasterRCNN variants (from fastest to most powerful):
      - fasterrcnn_mobilenet_v3_large_fpn
      - fasterrcnn_resnet50_fpn
      - fasterrcnn_resnet50_fpn_v2

    Args:
        variant: The FasterRCNN variant to use (see all options at
            https://pytorch.org/vision/stable/models.html#object-detection).
        pretrained: Whether to load model weights pretrained on COCO
        box_score_thresh: during inference, only return proposals with a classification
            score greater than box_score_thresh
    """

    def __init__(
        self,
        freeze_bn_stats: bool = False,
        freeze_bn_weights: bool = False,
        variant: str = "fasterrcnn_mobilenet_v3_large_fpn",
        pretrained: bool = False,
        box_score_thresh: float = 0.01,
    ) -> None:
        if not variant.lower().startswith("fasterrcnn"):
            raise ValueError(
                "The version must start with `fasterrcnn`. See available models at "
                "https://pytorch.org/vision/stable/models.html#object-detection"
            )

        super().__init__(
            model=variant,
            weights=("COCO_V1" if pretrained else None),
            num_classes=None,
            freeze_bn_stats=freeze_bn_stats,
            freeze_bn_weights=freeze_bn_weights,
            box_score_thresh=box_score_thresh,
        )

        # Modify the base predictor to output the correct number of classes
        num_classes = 2
        in_features = self.model.roi_heads.box_predictor.cls_score.in_features
        self.model.roi_heads.box_predictor = detection.faster_rcnn.FastRCNNPredictor(in_features, num_classes)