vidtracker 0.1.4

Video Object Tracking Toolkit

VidTracker

VidTracker is a Python package for object tracking

---

Tracker

Tracker	Paper	Description
DFSTracker	Distribution Fields for Tracking (CVPR 2012)	Smooth histogram fields with spatial/feature domain convolution
MILTracker	Visual Tracking with Online Multiple Instance Learning (CVPR 2009)	Online boosting with Haar features
LKTracker	Based on Lucas-Kanade optical flow	Tracks feature points and estimates affine transforms

Dataset

You can download the original MILTrack dataset (e.g., cliffbar) from:
--> https://bbabenko.github.io/miltrack.html

Extract the dataset into data/input/ to match the folder structure shown below.

Project Structure

.
├── config.json
├── data
│   ├── input
│   │   ├── cliffbar
│   │   │   ├── cliffbar_frames.txt
│   │   │   ├── cliffbar_gt.txt
│   │   │   ├── cliffbar_MIL_TR*.txt
│   │   │   ├── imgs/
│   │   │       └── imgXXXXX.png
│   │   ├── cliffbar.zip
│   └── output/
│   │   ├── cliffbar
│   │   │   └── imgXXXXX.png
├── tests/
└── vidtracker/
    ├── cli.py
    ├── dfs.py
    ├── lk.py
    ├── mil.py
    ├── util.py
    └── video.py

---

Installation

From PyPI:

pip install vidtracker

From GitHub:

pip install git+https://github.com/keyneoei/vidtracker.git

Or locally:

git clone https://github.com/keyneoei/vidtracker.git
cd vidtracker
pip install .

CLI Usage

vidtracker --input=data/input/cliffbar/imgs --output=data/output/cliffbar --tracker=DFS --show_frames

or

python -m vidtracker.cli --input=data/input/cliffbar/imgs --output=data/output/cliffbar --tracker=DFS --show_frames

Usage Example (Python)

from vidtracker import DFSTracker, MILTracker, LKTracker

tracker_type = "DFS"    # or "MIL", "LK"
frame = ...             # read video frame
init_bbox = ...         # init bbox
cfg = ...               # configuration (box)

if tracker_type == "DFS":
    tracker = DFSTracker(frame, init_bbox, cfg)
elif tracker_type == "MIL":
    tracker = MILTracker(frame, init_bbox, cfg)
elif tracker_type == "LK":
    tracker = LKTracker(frame, init_bbox, cfg)

# Process subsequent frames
x, y, w, h, angle = tracker.process_frame(next_frame)

---

References

• DFS: Sevilla-Lara, L., Learned-Miller, E. (2012). Distribution Fields for Tracking. CVPR.
• MIL: Babenko, B., Yang, M.-H., & Belongie, S. (2009). Visual Tracking with Online Multiple Instance Learning. CVPR.
• LK: Lucas, B., & Kanade, T. (1981). An Iterative Image Registration Technique. DARPA.

License

This project is licensed under the MIT License.