Meshworks
Meshworks strives to make the 3D capture of real-world assets effortless and accessible to anyone, anywhere. We are primarily targeting e-commerce sellers, from artists, to fashion designers, to culinary professionals, creating a platform for them to create, customize, and share 3D models of their products through photogrammetry.
Team Members
| Member | Photo | |
|---|---|---|
| Jihee Hwang | jiheeh@stanford.edu http://jiheehwang.com |
 |
| Andrew Huang | ahuang98@stanford.edu |  |
| Aakanksha Saxena | asaxena1@stanford.edu |  |
| Grace Cheng | gcheng21@stanford.edu |  |
| Farhan Azam | fazam@stanford.edu |  |
Team Skills Matrix:
| Member | Hat | Technical Skills | Non-Technical Skills | Personal Traits | Desired Growth | Weaknesses |
|---|---|---|---|---|---|---|
| Jihee | Green | Unity, VR, graphics, AI/deep learning | visual design, presentation skills | creative, motivated | product management skills, collaborative development | time management |
| Andrew | Green | ML, parallel computing | Debugging, writing, presenting | Energetic, paranoid, fun | Product management, VR development, user experience | Time management, getting nerd sniped, often tired |
| Aakanksha | White/Blue | AI/deep learning, Scraping | Writing, VC, business development | outgoing, hustler | VR, Unity, Product Development | time management |
| Grace | Blue | Backend, AI/ML | presentation skills, planning/organization, time management | ambitious, reliable | VR development, product management | indecision |
| Farhan | White | Unity, VR, Web Development, Graphics | planning, writing | conscientious, methodical | public speaking, product management | communication |
Documentation
Our product
Team Worldsight is building an intuitive pipeline, Meshworks, for e-commerce sellers to capture, customize, and share 3D models of their products with their customers.
We are also developing a separate web platform experience to showcase the potential for building on top of the post-pipeline production. Our pipeline can be divided into 2 sections:
Photogrammetry Request Workflow
- The client is served the Meshworks Website from our webserver, which is a dedicated Amazon EC2 instance.
- The webserver uploads photos to an Amazon S3 bucket, which stores both the photos used to create the 3D object mesh and the 3D object mesh itself.
- The webserver stores relevant information for each photogrammetry request in a MongoDB Atlas server. It stores the name of the user, email, tags, links to the photos in the S3 bucket, and more.
- The photogrammetry server, which is another EC2 instance, watches the MongoDB database for updates. When a request is made, it obtains all the necessary information to create a 3D object mesh, including the S3 links.
- The photogrammetry server downloads all the photos for the request from the S3 bucket.
- Once the photogrammetry pipeline is finished, the 3D object mesh is uploaded to the S3 bucket in glb format.
Meshroom is an open source photogrammetry software which takes in 2D photos of an object and outputs a 3D mesh object. You can find the source code and documentation for meshroom here. Using the software and command line interface is tedious and unintuitive for people with less technical experience, so we are wrapping a convenient GUI around the software package, and creating python scripts that run the necessary commmands to create the 3D mesh. The python scripts that run meshroom can be found in the pipeline folder and the code for the GUI can be found in the software-demo folder. The GUI uses Tkinter, a Python GUI toolkit.
2. Mesh Editor
Once the 3D mesh object is created, it will be placed in a folder designated by the user. After opening the Meshworks application, the user is led to the edit mesh scene. The edit mesh scene lets the user open a file explorer UI in VR, which can be used to directly select the scanned mesh to load. The mesh is then sent directly to the editor scene, where users can rotate, scale, and translate meshes into the appropriate orientation. The code for this can be found in the treehacks directory.
Dependencies
We are using Unity Version 2019.3.1f1 and the current stable build of Meshroom. Meshroom is built on top of AliceVision and CUDA.
For best performance, we recommend using Python (>=3.5) and having an nVidia GPU.
Partnerships
The Facebook Edmod team had two primary deliverables they wanted from us winter quarter: (1) partnerships with historical sites that we could develop as Stanford students and (2) scans of these sites demonstrating the feasibility of using Realities without training. We spent dozens of hours securing official partnerships with historical sites in Italy, The Netherlands, and Spain. This was a project-specific goal that we worked on outside of traditional software development in order to fullfill the deliverable requested by Facebook and to have scans we could use in our MVP pipeline. For a full list of our partnerships, can be found here.
During spring quarter, due to unexpected developments on campus as a result of the pandemic, all team members will be working remotely from our home locations and our deliverables will be updated to reflect this.
Team Communication:
Email us at [worldsightvr@gmail.com]
Reach us on Slack at [worldsight.slack.com]
Resources
Check out our google drive: https://drive.google.com/drive/folders/1RZd6VUiSgIe9Pw3Y4k7hHcuKc-0eMxvx?usp=sharing
Tech Stack in Product Overview in Wiki: https://github.com/cs210/Worldsight/wiki/