AI 3D Defect Detection Model

I need an end-to-end AI solution that takes a reference 3D model (CAD/mesh) and a drone-captured video of the physical object, aligns the two, flags every deviation, and produces a clear, human-readable report.

Scope
• Input video: a drone will orbit the object and record standard 4K/30 fps footage.
• The model must auto-extract frames, reconstruct a usable 3D representation, register it against the reference model, then detect
– missing or extra parts
– cracks, dents, scratches, or other surface damage
– incorrect dimensions, shapes, or part positions.
• A PDF/JSON report should list each defect, its coordinates, and severity, plus annotated images for fast visual review.

Deliverables
– Fully trained model with all weights/checkpoints.
– Clean, documented source code (Python preferred; OpenCV, PyTorch/TensorFlow, Open3D welcome).
– Inference script or API that accepts a video file and reference model and outputs the defect report.
– A short guide showing dataset preparation, training, and deployment steps.

Acceptance Criteria
1. Detect ≥90 % of seeded defects on a test set I will provide.
2. False-positive rate ≤5 %.
3. End-to-end processing time under 5 minutes for a 2-minute 4K video on a modern GPU.

If you have proven experience in 3D reconstruction, point-cloud registration, or industrial visual inspection, I’m ready to review past work and discuss the data you’ll need to get started.