# Copyright 2022 AI Singapore
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""🎠Instance segmentation model for generating high-quality masks."""
from typing import Any, Dict
from peekingduck.pipeline.nodes.model.mask_rcnnv1 import mask_rcnn_model
from peekingduck.pipeline.nodes.abstract_node import AbstractNode
[docs]class Node(AbstractNode): # pylint: disable=too-few-public-methods
"""Initializes and uses Mask R-CNN to infer from an image frame.
The Mask-RCNN node is capable detecting objects and their respective masks
from 80 categories. The table of object categories can be found
:ref:`here <general-instance-segmentation-ids>`. The ``"r50-fpn"`` backbone is
used by default, and the ``"r101-fpn"`` for the ResNet 101 backbone variant can also
be chosen.
Inputs:
|img_data|
Outputs:
|bboxes_data|
|bbox_labels_data|
|bbox_scores_data|
|masks_data|
Configs:
model_type (:obj:`str`): **{"r50-fpn", "r101-fpn"}, default = "r50-fpn"**. |br|
Defines the type of backbones to be used.
weights_parent_dir (:obj:`Optional[str]`): **default = null**. |br|
Change the parent directory where weights will be stored by
replacing ``null`` with an absolute path to the desired directory.
min_size (:obj:`int`): **default = 800**. |br|
Minimum size of the image to be rescaled before feeding it to the
backbone.
max_size (:obj:`int`): **default = 1333**. |br|
Maximum size of the image to be rescaled before feeding it to the
backbone.
detect (:obj:`List[Union[int, string]]`): **default = [0]**. |br|
List of object class names or IDs to be detected. To detect all classes,
refer to the :ref:`tech note <general-instance-segmentation-ids>`.
max_num_detections: (:obj:`int`): **default = 100**. |br|
Maximum number of detections per image, for all classes.
iou_threshold (:obj:`float`): **[0, 1], default = 0.5**. |br|
Overlapping bounding boxes with Intersection over Union (IoU) above
the threshold will be discarded.
score_threshold (:obj:`float`): **[0, 1], default = 0.5**. |br|
Bounding boxes with classification score below the threshold will be discarded.
mask_threshold (:obj:`float`): **[0, 1], default = 0.5**. |br|
The confidence threshold for binarizing the masks' pixel values; determines whether an
object is detected at a particular pixel.
References:
Mask R-CNN: A conceptually simple, flexible, and general framework for object
instance segmentation.:
https://arxiv.org/abs/1703.06870
Inference code adapted from:
https://pytorch.org/vision/0.11/_modules/torchvision/models/detection/mask_rcnn.html
The weights for Mask-RCNN Model with ResNet50 FPN backbone were adapted from:
https://download.pytorch.org/models/maskrcnn_resnet50_fpn_coco-bf2d0c1e.pth
"""
def __init__(self, config: Dict[str, Any] = None, **kwargs: Any) -> None:
super().__init__(config, node_path=__name__, **kwargs)
self.model = mask_rcnn_model.MaskRCNNModel(self.config)
def run(self, inputs: Dict[str, Any]) -> Dict[str, Any]:
"""Reads `img` from `inputs` and return the bboxes and masks of the detect
objects.
The classes of objects to be detected can be specified through the
`detect` configuration option.
Args:
inputs (Dict): Inputs dictionary with the key `img`.
Returns:
(Dict): Outputs dictionary with the keys `bboxes`, `bbox_labels`,
`bbox_scores` and `masks`.
"""
bboxes, labels, scores, masks = self.model.predict(inputs["img"])
outputs = {
"bboxes": bboxes,
"bbox_labels": labels,
"bbox_scores": scores,
"masks": masks,
}
return outputs