High Detail 3D Reconstruction from 2D (Gaussian splatting or NeRF)
Budget: $5,000 – $10,000 USD
We are preparing to develop a program that utilizes Gaussian Splatting technology (leveraging open-source resources) to convert 2D data into 3D backgrounds needed for content shooting in an LED virtual studio.
Project Goals:
Creating 3D backgrounds for existing virtual studios requires significant time and cost. Our plan is to enable anyone to quickly and easily generate 3D environments using this program. Like this video: https://youtu.be/fuXbPijS0WI?si=6Sz6xsB3W29JAafY
Requirements:
Task analysis and development design
- GUI design
- Windows program development
- Utilize open-source for AI components
Documentation Requirements:
- Weekly progress reports need to be prepared and delivered.
- A user manual detailing methods to acquire high-quality 2D data (direction, angle, lighting, weather conditions, etc.) is required.
Development Environment/Language:
- Gaussian Splatting technology (leveraging open-source)
- The software needs to be developed to run on Windows OS.
- Please suggest specifications for a PC, including graphics card, to ensure smooth operation of the software.
Key Features:
- Data Upload Function: Function to upload 2D video data for 3D generation.
- Preprocessing: Convert uploaded 2D video data into a suitable format for processing, ensuring consistency and compatibility with the Gaussian Splatting algorithm through resizing, normalization, and - other transformations.
- 3D Data Generation (Gaussian Splatting): Apply Gaussian Splatting technology to each point of the generated point cloud, distributing attributes (color, depth, etc.) into the surrounding 3D space to create 3D data. (it can be done using NeRF. or SFM based dense point cloud + gaussian splatting)
- Postprocessing: Improve the quality of the generated 3D model through noise reduction, smoothing, or other post-processing tasks.
- Visualization: Visualization function for pre-inspection and analysis of the generated 3D model.
Data Export: Function to export the final 3D data in a suitable format for integration with other applications.
Deliverables:
Original source code
Reference Sites/Apps:
- LUMA AI
- VOLINGA
Please note these two sites are platforms offering similar services and should be used for reference only. What we are creating is an offline standalone program for use in a virtual studio, not an online platform.
Ideal candidates should have:
- Strong experience in 3D reconstruction
- Proficiency in tools such as Gaussian Splatting or NeRF
- A background in computer graphics research
Considerations for Application:
- It would be beneficial if the applying company has experience with similar projects.
- Weekly progress reports need to be prepared and delivered.
- A user manual detailing methods to acquire high-quality 2D video data (direction, angle, lighting, weather conditions, etc.) is required.
Project Goals:
Creating 3D backgrounds for existing virtual studios requires significant time and cost. Our plan is to enable anyone to quickly and easily generate 3D environments using this program. Like this video: https://youtu.be/fuXbPijS0WI?si=6Sz6xsB3W29JAafY
Requirements:
Task analysis and development design
- GUI design
- Windows program development
- Utilize open-source for AI components
Documentation Requirements:
- Weekly progress reports need to be prepared and delivered.
- A user manual detailing methods to acquire high-quality 2D data (direction, angle, lighting, weather conditions, etc.) is required.
Development Environment/Language:
- Gaussian Splatting technology (leveraging open-source)
- The software needs to be developed to run on Windows OS.
- Please suggest specifications for a PC, including graphics card, to ensure smooth operation of the software.
Key Features:
- Data Upload Function: Function to upload 2D video data for 3D generation.
- Preprocessing: Convert uploaded 2D video data into a suitable format for processing, ensuring consistency and compatibility with the Gaussian Splatting algorithm through resizing, normalization, and - other transformations.
- 3D Data Generation (Gaussian Splatting): Apply Gaussian Splatting technology to each point of the generated point cloud, distributing attributes (color, depth, etc.) into the surrounding 3D space to create 3D data. (it can be done using NeRF. or SFM based dense point cloud + gaussian splatting)
- Postprocessing: Improve the quality of the generated 3D model through noise reduction, smoothing, or other post-processing tasks.
- Visualization: Visualization function for pre-inspection and analysis of the generated 3D model.
Data Export: Function to export the final 3D data in a suitable format for integration with other applications.
Deliverables:
Original source code
Reference Sites/Apps:
- LUMA AI
- VOLINGA
Please note these two sites are platforms offering similar services and should be used for reference only. What we are creating is an offline standalone program for use in a virtual studio, not an online platform.
Ideal candidates should have:
- Strong experience in 3D reconstruction
- Proficiency in tools such as Gaussian Splatting or NeRF
- A background in computer graphics research
Considerations for Application:
- It would be beneficial if the applying company has experience with similar projects.
- Weekly progress reports need to be prepared and delivered.
- A user manual detailing methods to acquire high-quality 2D video data (direction, angle, lighting, weather conditions, etc.) is required.