Pymoth

Installation

pip install pymoth

Contents

DataSet(directory)

A Namespace for handling entire MOT datasets, specifically designed for MOTChallenge (MOT16 onwards)

Each sequence and its corresponding labels are loaded on PyTrack initialisation.

• directory: A string indicating the path to the dataset

Methods

summary(tab_size=2)

Prints a summary of the sequences loaded into the PyTrack Object

• tab_size: int: number of spaces per tab

get()

Return dict of the Namespeace

add(dictionary)

Update dict with a new dictionary

Examples

Given layout of the MOTChallenge dataset as follows:

    .
    └── MOTChallenge
         ├── MOT15 # seqinfo.ini files will have to be created manually
         └── MOT16
             ├── test
             │   ├── MOT16-01
             │   │   ├── det
             │   │   │   └── det.txt
             │   │   ├── img1
             │   │   │   ├── 000001.jpg
             │   │   │   ├── 000002.jpg
             │   │   │   └── ...
             │   │   └── seqinfo.ini
             │   └── ...
             └── train
                  ├── MOT16-02
                  │   ├── det
                  │   │   └── det.txt
                  │   ├── gt
                  │   │   └── gt.txt
                  │   ├── img1
                  │   │   ├── 000001.jpg
                  │   │   ├── 000002.jpg
                  │   │   └── ...
                  │   └── seqinfo.ini
                  └── ...

MOT16 would be loaded with:

tracks = DataSet("path/to/MOTChallenge/MOT16")

Ground truth data for MOT16-02 would be accessed with:

tracks.train.MOT16_02.gt

Ground truth data for MOT16-02 would be displayed as a video with:

tracks.train.MOT16_02.gt.show(draw=True, show_ids=True, width=2)

Sequence()

An object to store object states throughout a video sequence.

Attributes

• info: A Namespace containing information about the Sequence
• frames: A list of Frame objects

Methods

init_frames(info=None, n=None, img_dir=None)

load_frames(img_dir, label_paths, info)

new_frame(img_path=None)

set_frame_paths(img_dir)

set_frame_path(frame, path)

create_instance(frame, kwargs)

add_instance(frame, instance)

get_images(width=1, scale=1, draw=False, show_ids=False)

get_n_frames()

get_n_ids()

get_n_instances(id=None)

get_instances(id=None)

get_ids()

get_boxes(id=None)

get_rects(id=None)

get_xywh(id=None)

get_conf(id=None)

get_appearances(id=None, shape=None)

show(scale=1, width=1, draw=False, show_id=False)

stream(image_only=False, scale=1)

Frame

An object to store instances from single frame

Attributes

• index: the index in the frame in the sequence
• img_path: the path to the image of the frame
• instances: list of Instance objects

Methods

create_instance(kwargs)

add_instance(instance)

get_image(width=1, scale=1, draw=False, show_ids=False)

get_n_instances()

get_ids()

get_n_ids()

get_boxes()

get_xywh()

get_rects()

get_conf()

get_appearances(shape=None)

Instance

Instance(id_number=-1, img_path=None, frame_index=None, bounding_box=None, coordinates=None, conf=None, state=None, color=None)

• id_number: int: the unique identification number of the instance
• img_path: str: the path to the image that contains the instance
• frame_index: int: the index number of the frame that contains the instance
• bounding_box: np.array(1, 4): the bounding box of the instance (left, top, width, height)
• coordinates: np.array(1, 3): the world coordinates of the instance (...)
• conf: int: the detection confidence of the instance (default = -1)
• state: str: the human-readable state of the instance
• color: tuple: the color used when drawing the instance bounding box

Attributes

Private

• _bounding_box: np.array(1, 4): the bounding box of the instance (left, top, width, height)
• _coordinates: np.array(1, 3): the world coordinates of the instance (...)
• _id: int: the unique identification number of the instance

Public

• color: tuple: the color used when drawing the instance bounding box
• conf: int: the detection confidence of the instance (default = -1)
• frame_index: int: the index number of the frame that contains the instance
• img_path: str: the path to the image that contains the instance
• mode: str: human readable description of the instance mode ('bounding_box' or 'world_coordinates')
• state: str: the human-readable state of the instance

Methods

set_bounding_box(bounding_box)

Store the instance bounding box and set the instance mode

set_coordinates(coordinates)

Store the instance world coordinates and set the instance mode

set_id(id_number)

Sets the instance id. If the instance color is not already set, sets the instance color based on the instance id number

• id_number: int: the unique identification number of the instance

get_bounding_box()

Returns np.array: the instance bounding box (left, top, width, height)

get_rect()

Returns: np.array(1, 4): the instance rect (left, top, right, bottom)

get_state()

Returns str: the human-readable state of the instance

get_xywh()

Returns: np.array(1, 4): the bounding box defined by (left, top, width, height)

get_id()

Returns: int: the unique identification number of the instance

get_appearance(shape=None, keep_aspect=True)

Returns: np.array: the image of the instance

• shape: the required shape of the output image
• keep_aspect: bool: whether to keep the object aspect ratio or not when resizing

show(image=None, draw=False, width=1, scale=1, show_ids=False)

Returns: np.array: the frame image containing the instance

• image: np.array: the image on which to draw the instance
• draw: bool: whether or not to draw the object bounding box / world coordinates
• width: int: the line width of the instance bounding box
• scale: int: the scale of the drawing
• show_ids: bool: whether or not to draw the instance id number

Utils

iou()

iou2()

nms()