# 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.
"""🎯 Human detection and tracking model that balances the importance between
detection and re-ID tasks.
"""
from typing import Any, Dict, List, Optional
import numpy as np
from peekingduck.pipeline.nodes.abstract_node import AbstractNode
from peekingduck.pipeline.nodes.model.fairmotv1 import fairmot_model
[docs]class Node(AbstractNode): # pylint: disable=too-few-public-methods
"""Initializes and uses FairMOT tracking model to detect and track people
from the supplied image frame.
FairMOT is based on the anchor-free object detector CenterNet with
modifications to balance the importance between detection and
re-identification tasks in an object tracker.
Inputs:
|img_data|
Outputs:
|bboxes_data|
|bbox_labels_data|
|bbox_scores_data|
|obj_attrs_data|
:mod:`model.fairmot` produces the ``ids`` attribute which contains the
tracking IDs of the detections.
Configs:
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.
score_threshold (:obj:`float`): **default = 0.5**. |br|
Object confidence score threshold.
K (:obj:`int`): **default = 500**. |br|
Maximum number of objects output during the object detection stage.
min_box_area (:obj:`int`): **default = 100**. |br|
Minimum value for area of detected bounding box. Calculated by
width * height.
track_buffer (:obj:`int`): **default = 30**. |br|
Threshold to remove track if track is lost for more frames
than value.
input_size (:obj:`List[int]`): **default = [864, 480]**. |br|
Size (width, height) of the input image to the model. Raw
video/image frames will be resized to the ``input_size`` before
they are fed to the model.
References:
FairMOT: On the Fairness of Detection and Re-Identification in Multiple
Object Tracking
https://arxiv.org/abs/2004.01888
Model weights trained by:
https://github.com/ifzhang/FairMOT
"""
def __init__(self, config: Dict[str, Any] = None, **kwargs: Any) -> None:
super().__init__(config, node_path=__name__, **kwargs)
self._frame_rate = 30.0
self.model = fairmot_model.FairMOTModel(self.config, self._frame_rate)
def run(self, inputs: Dict[str, Any]) -> Dict[str, Any]:
"""Tracks objects from image.
Specifically for use with MOT evaluation, will attempt to get optional
input `mot_metadata` and recreate `FairMOTModel` with the appropriate
frame rate when necessary.
Args:
inputs (Dict[str, Any]): Dictionary with keys "img". When running
under MOT evaluation, contains "mot_metadata" key as well.
Returns:
(Dict[str, Any]): Dictionary containing:
- bboxes (List[np.ndarray]): Bounding boxes for tracked targets.
- bbox_labels (np.ndarray): Bounding box labels, hard coded as
"person".
- bbox_scores (List[float]): Detection confidence scores.
- obj_attrs (Dict[str, List[int]]): Tracking IDs, specifically for use
with `mot_evaluator`.
"""
metadata = inputs.get(
"mot_metadata", {"frame_rate": self._frame_rate, "reset_model": False}
)
frame_rate = metadata["frame_rate"]
reset_model = metadata["reset_model"]
if frame_rate != self._frame_rate or reset_model:
self._frame_rate = frame_rate
self._reset_model()
bboxes, bbox_scores, track_ids = self.model.predict(inputs["img"])
bbox_labels = np.array(["person"] * len(bboxes))
bboxes = np.clip(bboxes, 0, 1)
outputs = {
"bboxes": bboxes,
"bbox_labels": bbox_labels,
"bbox_scores": bbox_scores,
"obj_attrs": {"ids": track_ids},
}
return outputs
def _get_config_types(self) -> Dict[str, Any]:
"""Returns dictionary mapping the node's config keys to respective types."""
return {
"input_size": List[int],
"K": int,
"min_box_area": int,
"score_threshold": float,
"track_buffer": int,
"weights_parent_dir": Optional[str],
}
def _reset_model(self) -> None:
"""Creates a new instance of the FairMOT model with the frame rate
supplied by `mot_metadata`.
"""
self.logger.info(
f"Creating new model with frame rate: {self._frame_rate:.2f}..."
)
self.model = fairmot_model.FairMOTModel(self.config, self._frame_rate)