Computer vision task
Object detection datasets and models
Bounding-box datasets and detectors you can fork, train, and deploy.
Object detection is the computer-vision task of finding and classifying every object in an image with an axis-aligned bounding box. Each annotation records a class label plus the box position and size, so a detector answers both "what is in the image?" and "where is it?" - the foundation for counting, tracking, and downstream analytics.
Detection datasets are typically annotated in the COCO, YOLO, or Pascal VOC formats, all of which Pictograph exports natively. In the editor you can draw boxes by hand, or prompt SAM 3 with a word or an example box to auto-label every matching object, then export to whichever format your training stack expects.
The datasets below carry bounding-box annotations you can fork into your workspace and train on directly. Pictograph trains detectors with YOLOX and RF-DETR and packages them as ONNX for edge or cloud inference, with no separate export-and-upload step.
Datasets 4
Trained models 1
Object detection FAQ
What is an object detection dataset?
An object detection dataset is a collection of images in which every object of interest is labelled with a bounding box and a class name. A model trained on it learns to draw those boxes on new images, locating and classifying objects at the same time. Public examples are usually annotated in the COCO, YOLO, or Pascal VOC formats.
What annotation format does object detection use?
The most common formats are COCO (a single JSON file), YOLO (one text file per image), and Pascal VOC (XML). Pictograph exports all three, plus CVAT, Datumaro, and others, from the same annotated dataset, so you can train with whichever stack you prefer without re-labelling.
How do I train an object detection model on Pictograph?
Fork or upload a dataset that has bounding-box annotations, then start a training run and choose the YOLOX or RF-DETR pipeline. Pictograph runs the export, the GPU training, and the ONNX packaging for you, and serves the finished model through the API or as a downloadable file.
Can I auto-label objects instead of drawing every box?
Yes. Pictograph's SAM 3 integration lets you prompt with a word (for example "car") or an example box, and it labels every matching object in the image, which you then review and correct. It is far faster than drawing each box by hand.
Start building today
$5/mo free compute. No payment method required.