havan2605/depth-estimation-demo

Object Depth Comparison Pipeline

End-to-end pipeline that detects and segments objects in an image, estimates per-pixel depth, then reports the average depth of each object and the difference between them.

Pipeline:

1. 1.SAM3 — promptable concept segmentation (text prompt → boxes + masks per class, in one step)
2. 2.Depth Anything V2 Large — monocular depth estimation
3. 3.Average depth per mask
4. 4.JSON comparison (depth per object + pairwise difference)

Note: facebook/sam3 is a gated model on the Hugging Face Hub. The environment running this app needs an HF_TOKEN (or a prior huggingface-cli login) for an account that has accepted the SAM3 license, otherwise model download will fail.