Frame analyzers

Pick an analyzer, upload a video, freeze a frame, and get an annotated result. The trained YOLO model is far more accurate on small stones and debris; GPT-5 Vision is a zero-shot fallback.

Pistachio detector (Trained YOLO)

Purpose-trained object detector via Roboflow. Much higher accuracy on small stones and debris — recommended.

Pistachio detector (GPT-5 Vision)

Zero-shot detection using OpenAI GPT-5 Vision. Good for novel contaminants but less accurate on tiny objects.