MooVision

MooVision is a computer-vision pipeline for detecting cross-sucking behaviour in socially housed dairy calves from angled overhead pen video. The goal is to reduce the time and effort required for manual review/labeling by automatically producing candidate events, clipped video segments, and structured metadata for downstream analysis.

---

Project Overview

Cross-sucking (a calf sucking directed at various body parts of another calf) is a welfare concern in group-housed calves and is currently studied via manual labeling of long video recordings. This project builds a scalable workflow that:

• Takes raw pen videos as input
• Runs a baseline detector (pretrained YOLOv8) with simple logic on top (e.g., proximity/overlap + temporal persistence)
• Outputs predicted event windows and metadata (start/end time, confidence, pen, weaning stage, day)
• Optionally generates clipped videos for review and evaluation

Project layout

mooVision/
├── docs/                     # Markdown files compiled by MkDocs into your static site
│   ├── index.md              # The documentation homepage
│   ├── ...                   
├── scripts/                  # Executable pipeline source code directories
│   ├── read_data/            # Read the data 
│   ├── split_data/           # Split the data 
│   ├── preprocessing/        # Preprocess data
│   ├── training/.            # Training models
│   ├── run_models/           # Scripts to run models on testing set
│   │   └── baseline/         
│   │   └── seq_NMS/         
│   │   └── yolo/         
│   │   └── run_testing.py    # For running fine-tuned YOLO model and Seq-NMS on test data
│   └── evaluation/
│   └── clipping    
├── tests/                    # Robust test suite validating code integrity
│   └── ...        
├── utils/                    # Supportive pipeline utility scripts
│   ├── build_clip_index.py   # Synchronizes dataset manifestations
│   ├── clip_frames_mp4s.py   # Baseline target image slicing utility
│   ├── count_labelled_clips.py # Audits dataset representation balances
│   ├── count_mp4s.py         # Validates physical cluster storage arrays
│   ├── extract_frames.py     # Multiprocessed image unpacking and tracking overlay
│   └── get_video.py          # Programmatic OpenCV stream verification tools
├── scripts_sockeye/          # Bash scripts to run in Sockeye
│   ├── 01_setup.sh           
│   ├── 02_read_and_split.sh  
│   ├── 03_preprocessing.sh   
│   ├── 04_train_yolo.sh      
│   ├── 05_baseline_testing.sh
│   ├── 06_yolo_testing.sh    
│   ├── 07_baseline_eval.sh   
│   └── 08_yolo_evaluation.sh
├── config.py                 # Master configuration file containing path definitions
├── mkdocs.yml                # Configuration file defining MkDocs plugins and themes
├── pyproject.toml            # Project dependency definitions managed via uv
├── reports/                  # Project proposal and final report
├── Makefile                  # Makefile to run the whole pipeline locally
└── run_training_pipeline.sh  # For running the whole pipeline in Sockeye

Pipeline Diagram

---
config:
  layout: elk
  elk: {}
  theme: base
---
flowchart TB
    indexCSV["index.csv"] --> readClips["read_all_clips_index.py"]
    readClips --> processedIndex["Processed Index Files"]
    processedIndex --> splitting["splitting.py"]
    splitting --> trainingIndex["Training Index"] & testingIndex["Testing Index"]
    trainingIndex --> preprocessing["preprocessing.py\n(can vary per model)"]
    preprocessing --> frameData["Frame-by-Frame Data"]
    frameData --> training["training.py"]
    training --> modelChoice{"Choose Training Script"}
    modelChoice --> yolo["training_yolo.py"] & model2["training_model2.py\n(or training_modelX.py)"]
    yolo --> trainedModel["Trained Model"]
    model2 --> trainedModel
    trainedModel --> testing["Model Testing"]
    testingIndex --> testing
    testing --> metadata["Metadata Output"]
    metadata --> evaluation["evaluation.py"] & clipping["clipping.py"]
    evaluation --> metrics["Evaluation Metrics\n(Accuracy, Precision, Recall, F1, F2, etc.)"]
    clipping --> clippedVideo["Result Clipped Video"]

     indexCSV:::dataStyle
     readClips:::scriptStyle
     processedIndex:::dataStyle
     splitting:::scriptStyle
     trainingIndex:::dataStyle
     testingIndex:::dataStyle
     preprocessing:::scriptStyle
     frameData:::dataStyle
     training:::scriptStyle
     modelChoice:::scriptStyle
     yolo:::scriptStyle
     model2:::scriptStyle
     trainedModel:::scriptStyle
     testing:::scriptStyle
     metadata:::dataStyle
     evaluation:::scriptStyle
     clipping:::scriptStyle
     metrics:::outputStyle
     clippedVideo:::outputStyle
    classDef scriptStyle stroke:#6366f1,fill:#eef2ff
    classDef dataStyle stroke:#2dd4bf,fill:#f0fdfa
    classDef outputStyle stroke:#f59e0b,fill:#fff7ed
    style indexCSV stroke:#FF6D00,fill:#FFE0B2

For full documentation visit mkdocs.org.