Skip to main content

WilData Configuration Reference

Documentation for all WilData configuration files used in data management operations.

Configuration Files​

FileDetailed ReferencePurpose
import-config-example.yamlImport Config →Dataset import configuration
bulk-import-*.yamlImport Config →Bulk import configurations
roi-create-config.yamlROI Config →ROI dataset creation
bulk-roi-create-config.yamlROI Config →Bulk ROI creation
gps-update-config-example.yamlGPS Config →GPS metadata update
label_studio_config.xml—Label Studio interface

!!! tip For detailed field-by-field documentation, see the individual config reference pages linked above.

import-config-example.yaml​

Purpose: Configure dataset import with transformations

Location: wildata/configs/import-config-example.yaml

Complete Configuration​

# Required: Source Information
source_path: "D:/annotations/dataset.json"
source_format: "coco"  # coco, yolo, ls (Label Studio)
dataset_name: "my_dataset"

# Pipeline Configuration
root: "D:/data"
split_name: "train"  # train, val, test
enable_dvc: false
processing_mode: "batch"  # batch, streaming
track_with_dvc: false
bbox_tolerance: 5

# Label Studio Options (for ls format)
dotenv_path: ".env"
ls_xml_config: "configs/label_studio_config.xml"
ls_parse_config: false

# ROI Configuration
disable_roi: false
roi_config:
  random_roi_count: 2              # Background samples per image
  roi_box_size: 384                # ROI size (pixels)
  min_roi_size: 32                 # Minimum object size
  dark_threshold: 0.7              # Dark image threshold
  background_class: "background"
  save_format: "jpg"               # jpg, png
  quality: 95                      # JPEG quality
  sample_background: true          # Sample background regions

# Transformation Pipeline
transformations:
  # Bbox Clipping
  enable_bbox_clipping: true
  bbox_clipping:
    tolerance: 5                   # Pixels outside image allowed
    skip_invalid: false            # Skip invalid bboxes
  
  # Data Augmentation
  enable_augmentation: false
  augmentation:
    rotation_range: [-45, 45]     # Rotation degrees
    probability: 1.0               # Augmentation probability
    brightness_range: [-0.2, 0.4]
    scale: [1.0, 2.0]
    translate: [-0.1, 0.2]
    shear: [-5, 5]
    contrast_range: [-0.2, 0.4]
    noise_std: [0.01, 0.1]
    seed: 41
    num_transforms: 2              # Augmentations per image
  
  # Image Tiling
  enable_tiling: true
  tiling:
    tile_size: 800
    stride: 640                    # Tile stride
    min_visibility: 0.7            # Min object visibility
    max_negative_tiles_in_negative_image: 2
    negative_positive_ratio: 1.0
    dark_threshold: 0.7

bulk-import Configs​

Purpose: Configure batch import of multiple datasets

Files:

  • bulk-import-train.yaml
  • bulk-import-val.yaml
  • bulk-import-config-example.yaml

Configuration Format​

source_paths:
  - "D:/annotations/dataset1.json"
  - "D:/annotations/dataset2.json"
  - "D:/annotations/dataset3.json"

source_format: "coco"
root: "D:/data"
split_name: "train"

# Shared settings (same as import-config)
processing_mode: "batch"
bbox_tolerance: 5

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

roi-create-config.yaml​

Purpose: Configure ROI dataset creation

Location: wildata/configs/roi-create-config.yaml

Configuration​

source_path: "annotations.json"
source_format: "coco"
dataset_name: "roi_dataset"

root: "data"
split_name: "val"  # Usually val or test
bbox_tolerance: 5

roi_config:
  roi_box_size: 128              # ROI crop size
  min_roi_size: 32               # Min object size to extract
  random_roi_count: 10           # Background samples per image
  dark_threshold: 0.7
  background_class: "background"
  save_format: "jpg"
  quality: 95
  padding: 10                    # Padding around object
  sample_background: true
  
  # Advanced options
  aspect_ratio_range: [0.5, 2.0]  # Valid aspect ratios
  min_object_area: 32             # Min area (pixels²)
  
ls_xml_config: null
ls_parse_config: false
draw_original_bboxes: false

bulk-roi-create-config.yaml​

Purpose: Bulk ROI dataset creation

Configuration​

source_paths:
  - "dataset1.json"
  - "dataset2.json"

source_format: "coco"
split_name: "val"

roi_config:
  roi_box_size: 128
  random_roi_count: 5
  background_class: "background"

gps-update-config-example.yaml​

Purpose: Update image GPS from CSV

Location: wildata/configs/gps-update-config-example.yaml

Configuration​

image_folder: "D:/images/"
csv_path: "gps_coordinates.csv"
output_dir: "D:/images_with_gps/"

# CSV Parsing
skip_rows: 0
filename_col: "filename"
lat_col: "latitude"
lon_col: "longitude"
alt_col: "altitude"

# Options
overwrite_existing: false        # Overwrite existing GPS
create_backup: true              # Backup original files
validate_coordinates: true       # Validate GPS coordinates

CSV Format​

filename,latitude,longitude,altitude
image001.jpg,40.7128,-74.0060,10.5
image002.jpg,40.7589,-73.9851,15.2

label_studio_config.xml​

Purpose: Label Studio annotation interface configuration

Location: wildata/configs/label_studio_config.xml

Example Configuration​

<View>
  <Image name="image" value="$image"/>
  <RectangleLabels name="label" toName="image">
    <Label value="elephant" background="red"/>
    <Label value="giraffe" background="blue"/>
    <Label value="zebra" background="green"/>
    <Label value="buffalo" background="yellow"/>
  </RectangleLabels>
</View>

Configuration Examples​

Import COCO with Tiling​

source_path: "D:/coco/annotations.json"
source_format: "coco"
dataset_name: "wildlife_tiled"

root: "D:/data"
split_name: "train"

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

Import Label Studio​

source_path: "D:/label_studio/export.json"
source_format: "ls"
dataset_name: "annotated_data"

ls_xml_config: "configs/label_studio_config.xml"
ls_parse_config: true

roi_config:
  roi_box_size: 128
  random_roi_count: 5

Import with Full Pipeline​

source_path: "raw_annotations.json"
source_format: "coco"
dataset_name: "processed_dataset"

transformations:
  enable_bbox_clipping: true
  enable_tiling: true
  tiling:
    tile_size: 800
    stride: 640
  enable_augmentation: true
  augmentation:
    num_transforms: 2
    probability: 0.8

roi_config:
  roi_box_size: 384
  random_roi_count: 10

Best Practices​

  1. Use absolute paths for cross-platform compatibility
  2. Enable bbox_clipping to fix annotation errors
  3. Tile large images for better training
  4. Sample background ROIs for balanced datasets
  5. Version control configuration changes

Next Steps​