Examples¶

The examples/ directory contains ready-to-use scripts for dataset preparation, mixing, and training.

Dataset Preparation¶

These scripts download and structure public datasets into the format CIA expects.

COCO People¶

python run.py task=prepare_data data.base=coco

Downloads COCO people dataset and converts annotations to YOLO format with captions.

Flickr30K Entities¶

python run.py task=prepare_data data.base=flickr30k

Downloads Flickr30K, extracts person regions with bounding boxes and captions.

FER (Facial Emotion Recognition)¶

python run.py task=prepare_data data.base=fer_real

Downloads FER dataset from Kaggle. Requires ~/.kaggle/kaggle.json (see Kaggle API).

MOCS (Moving Objects in Construction Sites)¶

python run.py task=prepare_data data.base=mocs

Downloads the MOCS dataset with 13 construction-related classes.

Training¶

YOLOv8 Object Detection¶

python run.py task=train data.base=coco

Trains a YOLOv8n model on the mixed dataset. Logs to Weights & Biases.

InceptionV3 FER Classification¶

python run.py task=train data.base=fer

Trains an InceptionV3 classifier on the mixed FER dataset.

Mixing Datasets¶

Create a mixed real+synthetic dataset:

python run.py task=mix data.base=coco ml.augmentation_percent=0.25

Data Structure¶

All prepared datasets follow this layout:

data/
├── real/{dataset}/
│   ├── train/{images,labels,captions}/
│   ├── val/{images,labels,captions}/
│   └── test/{images,labels,captions}/
├── generated/{dataset}/{controlnet-model}/
│   ├── metadata.yaml
│   └── *.png
└── mixed/{dataset}/