README.md

0. Directory structure:

Hens Tracking

├── data
│   ├── frames
│   │   └── frame_0.jpg, frame_1.jpg, ..., frame_204.jpg
│   └── video
│       └──  video.mkv
│
├──  train
│   ├── images
│   │   └── first 185 images
│   └── labels
│
├──  val
│   ├── images
│   │   └── remaining 20 images
│   └── labels
│
├──  annotate.py
│
├──  get_stuff.py
│
├──  track.py
│
└── data.yaml

1. Annotated the frames using ultralytics auto annotator

Run python annotate.py

(See output annotate_output.txt)

This creates a folder under data named 'frames_auto_annotate_labels' and gives a .txt file for each frame containing the segmentations of the hens it detected.

2. Put the labels into a proper structure for training i.e. specify custom class (hen)

Run python get_stuff.py

This will put all the necesaary files under train/labels. Move the last 20 files to val/labels for validation in the next stripped.

3. The model is now trained using a pretrained yolov8x-seg

Ensure that the directory path is good

Run yolo task=segment mode=train model=yolov8x-seg data=data.yaml epochs=20 imgsz=640 in a terminal opened in the directory

(See output train_output.txt)

This will create a runs folder and all the metrics and model weights will be stored there.

4. Use the best weights as input model for tracking

Run python track.py

(See output track_output.txt, uncomment line no. 34 if you want to visualize the output alongside running)

The final tracking video would be saved under the directory.