Dataset Preparation Tutorial

Learn how to prepare datasets for training using WilData.

Overview

This tutorial covers importing, transforming, and exporting datasets for wildlife detection training.

Prerequisites

• WilData installed
• Annotated images (COCO, YOLO, or Label Studio format)

Step 1: Import Dataset

Option A: Using Config File

Create config.yaml:

source_path: "D:/annotations/dataset.json"
source_format: "coco"
dataset_name: "wildlife_train"
root: "D:/data"
split_name: "train"

transformations:
  enable_tiling: true
  tiling:
    tile_size: 800
    stride: 640
    min_visibility: 0.7

Run import:

cd wildata
wildata import-dataset --config config.yaml

Option B: Direct CLI

wildata import-dataset annotations.json \
    --format coco \
    --name wildlife_train \
    --enable-tiling \
    --tile-size 800

Step 2: Apply Transformations

Tiling for Large Images

transformations:
  enable_tiling: true
  tiling:
    tile_size: 800
    stride: 640
    min_visibility: 0.7

Bbox Clipping

transformations:
  enable_bbox_clipping: true
  bbox_clipping:
    tolerance: 5
    skip_invalid: false

Step 3: Create ROI Dataset

For classification training:

cd wildata
scripts\create-roi-dataset.bat

# Or with CLI
wildata create-roi-dataset --config configs/roi-create-config.yaml

Step 4: Visualize

# Launch FiftyOne
wildata visualize-dataset --dataset wildlife_train --split train

Step 5: Export for Training

# Export to YOLO format
wildata dataset export wildlife_train --format yolo --output exports/yolo

# Export to COCO
wildata dataset export wildlife_train --format coco --output exports/coco

Complete Example

from wildata import DataPipeline

# Initialize
pipeline = DataPipeline("data")

# Import with transformations
result = pipeline.import_dataset(
    source_path="annotations.json",
    source_format="coco",
    dataset_name="wildlife_train",
    transformations={
        "enable_tiling": True,
        "tiling": {
            "tile_size": 800,
            "stride": 640
        }
    }
)

# Export for training
pipeline.export_dataset("wildlife_train", "yolo")

---

Next Steps: - Model Training Tutorial - WilData Scripts Reference