Depth Scene Analyzer
AI-powered spatial scene understanding with object size estimation from single images
Overview
This project combines GroundingDINO, SAM, and Depth Anything V2 to estimate real-world object sizes from single images. Users provide a reference object with known dimensions to calibrate measurements of other objects in the scene.
Key Technologies
PythonPyTorchTransformersOpenCVStreamlit
AI Models: GroundingDINO • SAM • Depth Anything V2
Key Results
95%
Detection Success
±25-35%
Measurement Accuracy
<1s
Processing Time
Pipeline Results
Real example: Office scene with laptop detection
Depth Estimation
Depth Anything V2 depth map
Object Segmentation
SAM precise masks
Final Analysis
Combined visualization
Result: Laptop dimensions estimated with 30% accuracy using reference calibration.See detailed analysis →