Roboflow

Roboflow supports training with YOLOv5, YOLOv8, YOLOv11, YOLO-World, RF-DETR, and other popular architectures. It also supports uploading custom model weights and fine-tuning foundation models like Florence and SAM2 directly within the platform.

Roboflow provides an optimized Inference server Docker container tailored for NVIDIA Jetson hardware. It configures drivers, libraries, and binaries for the Jetson CPU and GPU architectures, supporting JetPack 6.2 and enabling air-gapped offline operation with cached model weights.

Roboflow Universe is the world's largest open source computer vision repository containing over 200 million images, 200,000 datasets, and 50,000 fine-tuned models. Developers can browse, fork, and use pre-trained models or datasets to accelerate their own computer vision projects.

The Workflow Builder is a visual pipeline editor that lets you chain multiple model steps, post-processing logic, and integrations into a single deployable pipeline. Workflows can combine object detection, classification, OCR, and custom Python blocks, then deploy to cloud or edge with one click.

Vision Events is a production monitoring feature that records, searches, and analyzes what deployed computer vision models observe. It provides a centralized, filterable history of detections with metadata, enabling teams to track specific classes, investigate edge cases, and audit system performance across device fleets.