This GitHub Repository contains the code relating to "Reconstructing Artifact-Free 3D Scenes of Coral Reefs" for CS231A Spring 2025. Refer to the PDF file titled "CS231A Project Final Report" for more information.
The original WaterSplatting code can be found here: https://github.com/water-splatting/water-splatting/tree/main
The original nerfstudio code can be found here: https://github.com/nerfstudio-project/nerfstudio/tree/main/nerfstudio
The following WaterSplatting files were altered:
- water_splatting.py
The following nerfstudio files were altered:
- base_dataset.py
- full_images_datamanager.py
- render.py
- trainer.py
- viewer.py
The following files were created for this project:
- create_dataset.py: extracts frames one second apart from input videos
- run_colmap.py: preprocesses the data with COLMAP
- comparison_frames.py: produces comparison frames to visualize the outputs
The SeaThruNeRF dataset can be found here: https://drive.google.com/uc?export=download&id=1RzojBFvBWjUUhuJb95xJPSNP3nJwZWaT
The NOAA Dataset (sectioned by dive site) can be found here: https://drive.google.com/file/d/1K6Fnljo5O1r-hcqW89KPIGSdUf7ZUqez/view?usp=drive_link
The original website containing the NOAA videos can be found here: https://www.ncei.noaa.gov/access/metadata/landing-page/bin/iso?id=gov.noaa.nodc:0172858
MS7T2_semantic_segmentation_subset.ndjson is a small semantic segmentation dataset (subset of 50 images from MS7T2) that can be used in future work.
The Comparison Frames folder contains comparison frames of MS7T2, OS3T2N, and PS1T1 with WaterSplatting (labeled with "OG") and WaterSplatting with RAFT. A comparison frame of SeaThruNeRF-IUI3-RedSea is also provided.