Python package for multiple object tracking research with focus on laboratory animals tracking.

Last update: Sep 05, 2022

Related tags

Deep Learning motutils

Overview

motutils is a Python package for multiple object tracking research with focus on laboratory animals tracking.

Features

loads:
- MOTChallenge CSV,
- sleap analysis,
- toxtrac,
- idtracker,
- idtrackerai.
saves: MOTChallenge CSV
Mot, BboxMot and PoseMot classes backed by xarray dataset with frame and id coordinates
export to Pandas DataFrame
oracle detector: fake all knowing detector based on ground truth with configurable inaccuracies
different classes of tracked objects: point, bounding box, pose
interpolation of missing positions
find mapping between MOT results and ground truth
visualization:
- tracked positions / objects overlaid on a video
- montage of multiple videos with results and/or ground truth
cli
- visualization
- evaluation ()
- mot format conversion

Video comparison of multiple tracking methods and the ground truth.

Installation

pip install git+https://github.com/smidm/motutils

Usage

$ motutils --help
Usage: motutils [OPTIONS] COMMAND [ARGS]...

Options:
--load-mot FILENAME             load a MOT challenge csv file(s)
--load-gt FILENAME              load ground truth from a MOT challenge csv
file
--load-idtracker FILENAME       load IdTracker trajectories (e.g.,
trajectories.txt)
--load-idtrackerai FILENAME     load idtracker.ai trajectories (e.g.,
trajectories_wo_gaps.npy)
--load-sleap-analysis FILENAME  load SLEAP analysis trajectories (exported
from sleap-label File -> Export Analysis
HDF5)
--load-toxtrac FILENAME         load ToxTracker trajectories (e.g.,
Tracking_0.txt)
--toxtrac-topleft-xy 
   
    ...
position of the arena top left corner, see
first tuple in the Arena line in Stats_1.txt
--help                          Show this message and exit.

Commands:
convert    Convert any format to MOT Challenge format.
eval       Evaluate a single MOT file against the ground truth.
visualize  Visualize MOT file(s) overlaid on a video.

$ motutils convert --help

Usage: motutils convert [OPTIONS] OUTPUT_MOT

  Convert any format to MOT Challenge format.

$ motutils eval --help

Usage: motutils eval [OPTIONS]

  Evaluate a single MOT file against the ground truth.

Options:
  --write-eval FILENAME  write evaluation results as a CSV file
  --keypoint INTEGER     keypoint to use when evaluating pose MOT results
                         against point ground truth

$ motutils visualize --help

Usage: motutils visualize [OPTIONS] VIDEO_IN VIDEO_OUT
                          [SOURCE_DISPLAY_NAME]...

  Visualize MOT file(s) overlaid on a video.

Options:
  --limit-duration INTEGER  visualization duration limit in s
  --help                    Show this message and exit.

Python API Quickstart

>> mot.ds 
Dimensions: (frame: 4500, id: 5) Coordinates: * frame (frame) int64 0 1 2 3 4 5 6 ... 4494 4495 4496 4497 4498 4499 * id (id) int64 1 2 3 4 5 Data variables: x (frame, id) float64 434.5 277.7 179.2 ... 185.3 138.6 420.2 y (frame, id) float64 279.0 293.6 407.9 ... 393.3 387.2 294.7 width (frame, id) float64 nan nan nan nan nan ... nan nan nan nan nan height (frame, id) float64 nan nan nan nan nan ... nan nan nan nan nan confidence (frame, id) float64 1.0 1.0 1.0 1.0 1.0 ... 1.0 1.0 1.0 1.0 1.0 >>> mot.num_ids() 5 >>> mot.count_missing() 0 >>> mot.get_object(frame=1, obj_id=2)

Dimensions: () Coordinates: frame int64 1 id int64 2 Data variables: x float64 278.2 y float64 293.7 width float64 nan height float64 nan confidence float64 1.0 >>> mot.match_xy(frame=1, xy=(300, 300), maximal_match_distance=40)

Dimensions: () Coordinates: frame int64 1 id int64 2 Data variables: x float64 278.2 y float64 293.7 width float64 nan height float64 nan confidence float64 1.0 >>> mot.to_dataframe() frame id x y width height confidence 0 1 1 434.5 279.0 -1.0 -1.0 1.0 1 1 2 277.7 293.6 -1.0 -1.0 1.0 2 1 3 179.2 407.9 -1.0 -1.0 1.0 3 1 4 180.0 430.0 -1.0 -1.0 1.0 4 1 5 155.0 397.0 -1.0 -1.0 1.0 ... .. ... ... ... ... ... 22495 4500 1 90.3 341.9 -1.0 -1.0 1.0 22496 4500 2 187.9 431.9 -1.0 -1.0 1.0 22497 4500 3 185.3 393.3 -1.0 -1.0 1.0 22498 4500 4 138.6 387.2 -1.0 -1.0 1.0 22499 4500 5 420.2 294.7 -1.0 -1.0 1.0 [22500 rows x 7 columns]">
>>> from motutils import Mot
>>> mot = Mot("tests/data/Sowbug3_cut.csv")

>>> mot.ds
<xarray.Dataset>
Dimensions:     (frame: 4500, id: 5)
Coordinates:
  * frame       (frame) int64 0 1 2 3 4 5 6 ... 4494 4495 4496 4497 4498 4499
  * id          (id) int64 1 2 3 4 5
Data variables:
    x           (frame, id) float64 434.5 277.7 179.2 ... 185.3 138.6 420.2
    y           (frame, id) float64 279.0 293.6 407.9 ... 393.3 387.2 294.7
    width       (frame, id) float64 nan nan nan nan nan ... nan nan nan nan nan
    height      (frame, id) float64 nan nan nan nan nan ... nan nan nan nan nan
    confidence  (frame, id) float64 1.0 1.0 1.0 1.0 1.0 ... 1.0 1.0 1.0 1.0 1.0

>>> mot.num_ids()
5

>>> mot.count_missing()
0

>>> mot.get_object(frame=1, obj_id=2)
<xarray.Dataset>
Dimensions:     ()
Coordinates:
    frame       int64 1
    id          int64 2
Data variables:
    x           float64 278.2
    y           float64 293.7
    width       float64 nan
    height      float64 nan
    confidence  float64 1.0

>>> mot.match_xy(frame=1, xy=(300, 300), maximal_match_distance=40)
<xarray.Dataset>
Dimensions:     ()
Coordinates:
    frame       int64 1
    id          int64 2
Data variables:
    x           float64 278.2
    y           float64 293.7
    width       float64 nan
    height      float64 nan
    confidence  float64 1.0

>>> mot.to_dataframe()
       frame  id      x      y  width  height  confidence
0          1   1  434.5  279.0   -1.0    -1.0         1.0
1          1   2  277.7  293.6   -1.0    -1.0         1.0
2          1   3  179.2  407.9   -1.0    -1.0         1.0
3          1   4  180.0  430.0   -1.0    -1.0         1.0
4          1   5  155.0  397.0   -1.0    -1.0         1.0
      ...  ..    ...    ...    ...     ...         ...
22495   4500   1   90.3  341.9   -1.0    -1.0         1.0
22496   4500   2  187.9  431.9   -1.0    -1.0         1.0
22497   4500   3  185.3  393.3   -1.0    -1.0         1.0
22498   4500   4  138.6  387.2   -1.0    -1.0         1.0
22499   4500   5  420.2  294.7   -1.0    -1.0         1.0
[22500 rows x 7 columns]

Documentation

See the quickstart and tests for now.

Write me if you would like to use the package, but the lack of documentation is hindering you. You can easily reorder my priorities on this simply just by letting me know that there is an interest.

Python package for multiple object tracking research with focus on laboratory animals tracking.

Related tags

Overview

Features

Installation

Usage

Python API Quickstart

Documentation

Owner

Matěj Šmíd

The first machine learning framework that encourages learning ML concepts instead of memorizing class functions.

Merlion: A Machine Learning Framework for Time Series Intelligence

Framework for joint representation learning, evaluation through multimodal registration and comparison with image translation based approaches

Use AI to generate a optimized stock portfolio

(AAAI2020)Grapy-ML: Graph Pyramid Mutual Learning for Cross-dataset Human Parsing

This program can detect your face and add an Christams hat on the top of your head

Weakly supervised medical named entity classification

HAT: Hierarchical Aggregation Transformers for Person Re-identification

code for CVPR paper Zero-shot Instance Segmentation

zeus is a Python implementation of the Ensemble Slice Sampling method.

This program automatically runs Python code copied in clipboard

TorchX: A PyTorch Extension Library for More Efficient Deep Learning

Drone-based Joint Density Map Estimation, Localization and Tracking with Space-Time Multi-Scale Attention Network

A curated list of awesome resources combining Transformers with Neural Architecture Search

Automatic Idiomatic Expression Detection

Official implementation of the paper Do pedestrians pay attention? Eye contact detection for autonomous driving

NPBG++: Accelerating Neural Point-Based Graphics

🔥🔥High-Performance Face Recognition Library on PaddlePaddle & PyTorch🔥🔥

This initial strategy was developed specifically for larger pools and is based on taking a moving average and deriving Bollinger Bands to create a projected active liquidity range.

A self-supervised 3D representation learning framework named viewpoint bottleneck.