I connected an LLM to SAM 3
Автор: Applied Tensors
Загружено: 2025-12-06
Просмотров: 1273
Описание:
Large language models don't have eyes. When you ask Claude or GPT to count objects in an image, it's not counting anything. It's guessing based on pattern matching. Sometimes it's right. Sometimes it's confidently wrong.
LINK TO THE PROJECT ON GITHUB - https://github.com/Tylerbryy/iris
IRIS (Iterative Reasoning with Image Segmentation) takes a different approach. Instead of letting Claude guess, it forces verification through Meta's SAM3 segmentation model. Ask "is this car running a red light?" and Claude doesn't hallucinate an answer. It segments the red light. Gets coordinates. Segments the car. Analyzes the spatial relationship. Returns an answer grounded in actual visual evidence.
Повторяем попытку...
Доступные форматы для скачивания:
Скачать видео
-
Информация по загрузке:
