open-edge-platform/geti
Build computer vision models in a fraction of the time and with less data.
Task: Computer vision – image classification, object detection, instance segmentation.
Dataset: User‑provided image datasets; supports COCO, Pascal VOC, YOLO and a Geti‑native format.
Architecture: Includes RF‑DETR, DINOv3 DETR, YOLO26, YOLOX, D‑FINE, Mask R‑CNN, RT‑DETR, YOLO11/12, YOLO26, EfficientNet, ViT, DINOv2, MobileNetV3, etc.
Input: Raw images (or video frames) supplied via the UI or Python API.
Output: Trained model (exported as OpenVINO IR/ONNX) and predictions (bounding boxes, masks, class labels).
Training: Python getitune engine – engine.train(max_epochs=…); also via Geti GUI with annotate‑train‑predict loop.
Loss: Not specified in the README.
Benchmark: Not specified in the README.
Additional: Models are automatically exported with OpenVINO™ for CPU, Intel® XPU (Arc GPUs, Core Ultra) or NVIDIA® CUDA; includes INT8 quantization for edge deployment.