Elephant-BoTSORT.mp4
Elephant-ByteTrack.mp4
Download and Unzip this repository. Navigate to the repository in terminal and type the following:
conda create -n yolo_env python==3.8
conda activate yolo_env
conda install pytorch torchvision torchaudio -c pytorch
pip install -U ultralytics wandb
pip install supervision==0.3
pip install selenium
These are all the dependencies of the project, navigate to 1_Data_Collection
and start running data_collection.py
.
The Wildlife Tracker Project is an Object Tracker customized for tracking Wildlife in India. This project is part of a collaborative effort with Weights and Biases to create a beginner friendly introduction to YOLOv8.
It detects and tracks six classes of animals:
Class | Animal |
---|---|
0 | Tiger 🐯 |
1 | Elephant 🐘 |
2 | Rhinoceros 🦏 |
3 | Bison 🦬 |
4 | Leopard 🐆 |
5 | Lion 🦁 |
This project guides you through every step of Object Tracking Pipeline. The pipeline is discuessed briefly:
- Data Collection: In the Data Collection step, we use Selenium library to Automate WebScraping and saving animal images.
- Data Annotation: We use ModifiedOpenLabeling (by Ivan Goncharov) to Annotate Images and save it using Custom Train Test Split.
- Experiment Tracking with W&B: We upload our dataset as
wandb.Artifact
and visualize our image along with the bounding boxes. Explained in-depth in the blog post (Linked below). - Custom Training YOLOv8: We train YOLOv8 with our Scraped Data. We train and log metrics to
wandb
- Custom Tracking with YOLOv8: We use the native tracking support provided by
ultralytics
and track with two SOTA tracking algorithms :BoTSORT
andByteTrack
.
It also has interactive exercises to keep you engaged!
Link to the blog post: Blog post is almost done!
Link to the video: Video will be out soon!
The assets in the Assets
folder are all from this amazing free stock photos website called Pexels.
Here are image credits (in no particular order):
- Elephant Grasslands - Video by P'MA'
- Lion - Photo by Gary Whyte
- Elephants Crossing - Video by Rihan Bezuidenhout from Pexels
- Bird - Photo by Philippe Donn
- DogVid - Video by Free Videos
- Apples - Photo by Josh Hild
Also, a huge thanks to the following people and resources:
- Ivan Goncharov - ModifiedOpenLabeling
- Piotr Skalski and RoboFlow - Supervision
If you have read this far, consider starring the project to show some love. Reach out if you have any questions.
My email: mukilan.git@gmail.com