Instants-ngp Output Issue After Reality Scan and Colmap

[navegive]

Hello,

I am a student currently studying Instants-ngp.

I first aligned my images in Reality Scan, then exported them to Colmap, and finally tried running them in Instants-ngp. However, as shown in the attached photo, I only see white clouds and cannot see the object I photographed.

For reference, I prepared my dataset using a DSLR and drone, resulting in 2,133 PNG images. Since the dataset was quite large, I downscaled the images (by 1/2) when exporting from Reality Scan.

My computer specs are: AMD 5950X, RTX A6000, and 64GB of RAM.

Could you please let me know what I might have done wrong? I would really appreciate your help.

Thank you!

[navegive]

I'm following up on my previous post regarding the "white clouds" output issue. I hope my last question wasn't a bother.

I have an update that might shed some light on the problem. Based on some advice I found online, I decided to try processing a much smaller dataset. I used only 200 images from the original set, and this time, the process completed without any issues. The result looks good, and the object is properly reconstructed, unlike the "white clouds" output I had before.

This makes me curious about something. I've seen many other NeRF projects and tutorials that successfully use thousands of images. My initial issue seems to have been resolved by drastically reducing the number of photos.

This leads me to a new question: Was my original problem (the 'white clouds' and incomplete object) caused by simply having too many images in my dataset? Is there a recommended number of images for a NeRF dataset?

Any insights on the trade-off between dataset size and quality for NeRF would be greatly appreciated.

Thank you again for your help.

P.S. Below are captured images from Colmap and Instant-NGP using the 2133-image dataset.