---
tags: application, 2022
---

# Statues Live #

Traditional photogrammetry suffers from several problems in close-range applications like cultural heritage, most noticeably speed, pattern reconstruction, reflective surfaces. Deep learning (popularly known as AI), has revolutionized many fields (...) and computer vision (...) has done so for photogrammetry.

Additionally, traditional photogrammetry predominantly relies on closed source proprietary software, while DL can be provided within a fully open source framework.

## Solution ##

- _3d rekonstruktion_, det er meget stort felt i computer vision og jeg har løbet et par projekter igennem, der kan tilpasses dine data: https://github.com/natowi/3D-Reconstruction-with-Deep-Learning-Methods
- _4d information stack rekonstruktion_, på baggrund af 1 skal vi rekonstruere statuers decay. det er en normal prædiktionsopgave, hvor input er din 3d model og output er eksempler på (måske simuleret) decay. jeg vil tro, der allerede eksisterer data til dette.

low resource challenges are solved with data augmentation of statues and transfer learning from existing 3d object libraries

R&D: train deep sttslv-conv for 3d reconstruction and 4d information stack reconstruction

VRE: virtual research environment for StatueLive participants and stakeholder
* training of models in CUDA and production-friendly TensorFlow
* dedicated StatuesLive APP (containerized application) provides intuitive interface (Jupyter) for solving inference problems (modeling statue decay based on 4d model) enabeling interactive experimentation with/-out code
* Tutorials to facilitate both retraining and inference

OS & RDM:
* minimal implementation of FAIR-by-design (DCAT, json)
* all code and models are provided under the OSI-approved MIT license
* tutorials are provided under the CC BY 4.0

## Improvements ##

- speed (når 1 er trænet, kan nye rekonstruktioner laves meget hurtigt)
- repeathed pattern reconstruction
- glossy and reflective surfaces