Capture Utility

FFmpeg

FFmpeg is the standard command-line media toolkit used in 3DGS pipelines to extract still frames from video before alignment and training.

Stage: UtilityVideo capturesBatch processingDataset prep

Open official linkFFmpeg Website->

What It Does

Many practical 3DGS captures start as video because walking around a scene is faster than taking hundreds of stills. FFmpeg turns that footage into a controlled image sequence.

The trick is extracting enough frames for overlap without flooding COLMAP with near-duplicates that slow matching and add blur.

How To Use It In 3DGS

Record slow, stable video with locked exposure and focus when possible.
Extract frames at a modest rate such as 1-3 fps for slow walkthroughs, then adjust for scene speed.
Optionally resize frames to keep alignment and training practical.
Send the frame folder to COLMAP, Nerfstudio, Postshot, or another downstream tool.

Things To Watch

Too many near-identical frames can make alignment slower without improving quality.
Motion blur and rolling shutter artifacts harm feature matching.
Phone rotation metadata can surprise downstream tools, so inspect extracted images before processing.

Official And Related Links

FFmpeg Website FFmpeg Documentation Image2 Muxer Documentation

Practical Extraction Patterns

For a slow walkthrough, start around 2 fps and inspect overlap before extracting a huge set.
Use sequential filenames such as `frame_%05d.jpg` so COLMAP and manual review stay predictable.
If the video is 4K or higher, consider a test alignment at reduced width before committing to full-resolution training.

FFmpeg Formats: image2 FFmpeg Filters