MLOps 3D Scene Reconstruction
Getting Started
Installation Guide
Prerequisites
Step 1 — Clone the Repository
Step 2 — Download the Dataset
Docker Setup (Recommended)
Step 3a — Configure the Environment File
Step 3b — Generate Docker Secrets
Step 3c — Generate TLS Certificates
Step 3d — Clone External Dependencies
Step 3e — Launch the Full Stack
Native Python Setup (Developer Mode)
Step 3a — Build ASMK
Step 3b — Build CroCo / DUSt3R Kernels
Step 3c — Build Remaining Packages
Step 3d — Create the Python Virtual Environment
Pre-trained Model Weights
Verifying the Installation
Troubleshooting Installation
Usage Guide
Running the System
Docker (Production)
Native (Development)
Running the Offline DVC Pipeline
Uploading Data via the UI
API Authentication
Uploading Data via the API
Polling Job Status
Downloading Results
Checking Data Drift
Triggering Retraining
Example End-to-End Workflow
Disabling Authentication (Development Only)
User Interface
UI Guide
Overview of the Interface
Step 1 — Logging In
Step 2 — The Header
Step 3 — The Upload Panel (Left Panel, Top)
Step 4 — The Stage Tracker (Left Panel, Middle)
Step 5 — The 3D Model Viewer (Right Panel)
Step 6 — The Stats Table (Bottom Panel)
Step 7 — Downloading Results
Step 8 — Starting a New Reconstruction
Drift Warning Banner
Error States
Frequently Asked Questions (UI)
System Reference
System Architecture
High-Level Architecture
Frontend
API Gateway and GPU Worker (Ray Serve)
Offline MLOps Pipeline
Monitoring Stack
Data Flow
Networking
Security Boundaries
Pipeline Documentation
Pipeline Overview
Stage 1 — Data Validation
Stage 2 — Exploratory Data Analysis and Baselines
Stage 3 — Image Preprocessing
Stage 4 — Data Preparation
Stage 5 — Scene Reconstruction (Core Pipeline)
Shortlist Generation
Feature Extraction and Matching
COLMAP Incremental SfM
Stage 6 — Evaluation
Online Inference Pipeline
Model Selection and Promotion
Drift Monitoring and Retraining
API Reference
Authentication
Endpoint Reference
POST /auth/token
GET /health
GET /ready
GET /metrics
POST /upload
GET /status/{job_id} | GET /jobs/{job_id}
GET /download/jobs/{job_id}
GET /download/jobs/{job_id}/csv
GET /download/jobs/{job_id}/{filename}
GET /clusters/{job_id}
GET /jobs/{job_id}/insights
POST /drift
POST /drift/trigger-retrain
Error Responses
MLOps & Operations
Security & Compliance
Overview of Security Controls
1. Docker Secrets Management
2. JWT Bearer Token Authentication
3. Access Logging
4. TLS 1.3 (Transport Security)
5. CI/CD Security Scanning
Protected API Endpoints
Alertmanager Authentication
Environment Variables
Security Recommendations for Production
Data Sources
Dataset Summary
IMC 2025 Training Dataset
IMC 2025 Test Dataset
Data Versioning
Preprocessing Assumptions
Known Biases and Limitations
Project Context
Business Understanding
Problem Statement
Use Cases
ML Metric
Business and Operational Metrics
Data Source
Stakeholders
Definition of Done
Frequently Asked Questions
General
Installation & Setup
Using the API
Reconstruction Quality
MLOps / DVC / MLflow
Monitoring & Alerts
Data & Drift
MLOps 3D Scene Reconstruction
Index
Index