search

SharkWipf / nerf_dataset_preprocessing_helper

Intelligent(?) image sharpness filtering for optimized NeRF training datasets
MIT License
62 stars 4 forks source link

dataset #3

Open Harry-KIT opened 6 months ago

Harry-KIT commented 6 months ago

Hi @SharkWipf. Thank you for repo! Could you share sample dataset to check your code?

SharkWipf commented 6 months ago

The data I used was just a couple of videos of my house, no proper datasets. I wrote this because I needed to improve my own NeRFs primarily, not to have academic-level reproducible results (there is no whitepaper linked to any of this or anything).

There is really only a single operation these scripts execute though: They take every frame and run a standard OpenCV Laplacian variance check over it, which is just a way of detecting image sharpness. The rest is all just filtering logic to select the best frames, based on that output.

Any input video or frames should give you the same results, I don't think the videos I took of my house (which, sorry, I'm not sharing) are going to make a difference.

This script really just is a really simple way to automate laplacian variance, nothing more.