Only Three Files Contain the SFM Mainly(sfm_only.py, vid.mp4, output_sfm_cloud.ply) and all other files are beliong to COLMAP(better version but based on SFM)
This repository presents my exploration of Structure from Motion (SfM) — a technique in Computer Vision used to estimate the 3D structure of a scene from a set of 2D images or a video sequence.
The project is divided into two major parts:
- Custom SfM Implementation (from Scratch) →
sfm_only.py- Built using OpenCV, NumPy, and SciPy.
- Produces a colored point cloud file
output_sfm_colored.ply.
- COLMAP-based Reconstruction (for comparison)
- Demonstrates how professional photogrammetry software performs the same pipeline automatically.
- Produces dense point clouds and meshed 3D models.
Structure from Motion (SfM) is a computer vision pipeline that takes multiple overlapping images (or frames from a video) and reconstructs:
- The 3D geometry of the scene
- The camera motion (position and orientation)
Essentially, SfM transforms a regular phone video into a 3D point cloud that can be visualized using software like MeshLab or CloudCompare.
| Component | File | Description |
|---|---|---|
| 🧠 My Custom SfM | sfm_only.py |
Python code implementing SfM from scratch (feature extraction, motion estimation, triangulation, bundle adjustment, and colored 3D reconstruction). |
| 🌈 Output (My SfM) | output_sfm_colored.ply |
Final colored 3D point cloud generated using my SfM pipeline. |
| 🧱 COLMAP Reconstruction | Other .py and .ply files |
Dense reconstruction and mesh generation using COLMAP commands. |
| 📸 Artifacts Folder | artifacts/ |
Contains reconstruction outputs (sparse cloud, dense model, mesh). |