specklesystems / SpeckleHackathon-ImgToPointCloud

Speckle ML demo app (image to pointcloud)

WARNING: This repo is the result of a 1day hackathon and is very unstructured.

Description of directories:

• python-connect: quick proof of concept way of sending pointclouds to a speckle server
• web-app: The frontend Vue app
• server: The final backend (includes a copy of the built frontend, for serving)

---

The server directory started with a clone of the ML model repository that can be found at https://github.com/ialhashim/DenseDepth

In there, there are some extra files that were added/modified:

• cherry.py - the startup script for this app
• run.py - edited from a previous script - the pre-processing and post-processing logic
• sp.py - utility to send pointclouds to a speckle server
• Dockerfile - for building a docker image with this app
• Makefile - build and run commands for easily build and run the docker image

To build the app:

• go into the server directory
• follow the instructions in the original README.md to download the NYU pretrained model.
• run make build
• edit Makefile to include your config
• run make run