Guides for visual AI teams.
Practical guides on annotation best practices, dataset formats, model training, and deployment — written for people building real systems.
Computer vision, in plain language.
The concepts behind every dataset-to-model workflow — no jargon.
Image annotation
Labeling images so a model can learn from them — drawing bounding boxes, outlining shapes (segmentation), or assigning a class. Annotation quality sets the ceiling on model quality.
Detection vs segmentation
Object detection puts a box around each object and names it. Segmentation outlines the exact shape (a mask). Use detection for “where and what”; use segmentation when precise boundaries matter, like a crack or a leaf.
Classification
Assigning one label — or several, for multi-label — to a whole image rather than locating objects within it. Useful for pass/fail and category tasks.
Dataset quality
Balanced classes, clean labels, representative samples, and correct train/validation/test splits. Leakage between splits silently inflates accuracy; Anastig runs health checks to catch it.
Human-in-the-loop
AI proposes labels; a person reviews, edits, or rejects them. You keep the speed of automation without letting silent mistakes into the dataset — the model is only as good as the labels you approve.
Privacy-preserving workflows
Doing as much as possible where the data already lives. In Anastig, annotation and export run in your browser, and heavy models run on a GPU worker your organization controls, so sensitive images can stay in your boundary.
Which export format when.
Anastig Studio exports each of these, validated before download. See the full format reference or the Export Center.
| Format | Files | Best for |
|---|---|---|
| YOLO / YOLO-seg | .txt + dataset.yaml | Fast detection and segmentation training across the YOLO family and many modern trainers. |
| COCO JSON | .json | Detection and segmentation with rich metadata; widely supported by research tooling. |
| Pascal VOC | .xml | Classic detection format — one human-readable XML per image. |
| Classification CSV | .csv | Simple image-to-label tables for classification tasks. |
In-depth guides.
Long-form tutorials are being written. Start with the essentials above or the interactive demo.
How to Label Images for AI
A complete guide to image annotation for computer vision models.
Guide coming soonObject Detection Annotation Guide
Annotating bounding boxes for YOLO and COCO object detection models.
Guide coming soonBuilding a Classification Dataset
How to create and structure image classification datasets for training.
Guide coming soonSegmentation Annotation Guide
Instance and semantic segmentation labeling techniques.
Guide coming soonYOLO Dataset Format Explained
Understanding YOLO label files, dataset.yaml, and directory structure.
Guide coming soonCOCO JSON Format Guide
COCO format annotations, instances, categories, and metadata.
Guide coming soonDataset Splitting Best Practices
Train/val/test splits, stratification, and avoiding data leakage.
Guide coming soonNegative Samples in Object Detection
When and how to include images without annotations in training datasets.
Guide coming soon