Machine Learning Art

How does 3D reconstruction work?

DPT + 3D Voxels — Vision Transformers DEMO

In Datasculpting, 3D reconstruction is the foundation of the work. The following article describes the 3D Voxel reconstruction technique based on depth estimation.

• April 2022 — AI art tools update can be found ➡️ HERE ⬅️

DPT + 3D Voxels

Learning-based techniques for 3D reconstruction have acquired a lot of traction in recent years due to their promising outcomes. In contrast to 2D pictures, however, 3D cannot be represented in its canonical form to make it computationally and memory-efficient. Furthermore, because to the limited data available from the image for 3D reconstruction, creating a 3D model straight from a single 2D image is considerably more difficult. The resolution, efficiency, and smoothness of 3D models necessary for many practical applications are still lacking in existing learning-based systems.

3D reconstruction is the technique of capturing the shape and look of natural things in computer vision.

dense prediction transformer

Convolutional networks are the foundation of nearly all extant dense prediction architectures. The dense prediction transformer, or DPT, is a neural network design that uses visual transformers to successfully perform dense prediction tasks. DPT architecture provides more fine-grained and globally coherent predictions.

3D Voxel is a value on a regular grid in three-dimensional space in 3D computer graphics. Like pixels in a 2D bitmap, Voxels do not usually have their location (or coordinates) explicitly stored with their values.

The best way to understand how the architecture described above works is to visit Radamés Ajna’s DEMO — the project page (scroll down)

AI art maker

It predicts the depth of an image using the DPT model and then reconstructs the 3D model as voxels. In this example I used my father’s bas-relief. This article is part of the project myFatherinthecloud.ai.

While depth estimation is not a new concept, Ajna’s demonstration has made them more accessible to everyone. Depth measurements vary a lot, and depth maps had a lot of “holes” when no readings were taken. The higher-level surface geometry must be deduced from this noisy point data to create 3D models for my project or applications such as games, physics, or CAD.

title: Dpt Depth Estimation + 3D Voxels
the author : Radamés Ajna

Project Page — DEMO :

Keywords: computer vision, Artificial Intelligence, datasets, Machine Learning, AI art, art, digital art, 3D Voxel, DPT, Pattern Recognition, Dense Prediction Transformer, zero-shot

I invite you to explore the concept of “AI creativity” by reading and learning from the many articles found on 🔵 MLearning.ai 🟠

• Check out my instagram with new material every week
• If you enjoyed this, follow me on Medium for more
• Want to collaborate? Let’s connect on LinkedIn
• https://linktr.ee/datasculptor
• 3D Machine Learning generated model on sketchfab

Data Scientists must think like an artist when finding a solution when creating a piece of code. Artists enjoy working on interesting problems, even if there is no obvious answer.

All our writers (members) receive the opportunity to be promoted on our social media, which increases the popularity of articles published on MLearning.ai

1. Linkedin (9.1K+ ML-professionals)
2. Twitter (4.8K+ followers)
3. Instagram (2.2K + followers )
4. Sketchfab * — individual vRooML!
5. Facebook
6. Youtube
7. Apple Podcasts
8. Substack

🔵 Submission Suggestions