lilac ingot
Hey all! I made Detect Anything, a tool that detects objects from any text prompt and outputs labeled data for YOLO training.
Examples: “cat’s left eye”, “girl carrying glasses”.
You can try it here: Detect Anything
Despite the high-quality output, it’s expensive and not suitable for real-time applications.
Looking for testing & feedback: does it generate usable YOLO datasets? Limitations? Ideas?
lilac ingot
No, it's definitely not possible for real-time video since it takes about 2–10 seconds per image and is too costly if called for every frame. What it is very useful for is generating labeled data to train your own YOLO model. It also works well for use cases where image analysis is non‑real‑time and/or when you can take advantage of parallel requests of up to 30 at a time.
prisma forum
Pretty cool, what are its limits?
warm oracle