AllenInstitute / segmentation-labeling-app

Data pipeline and UI for human labeling of putative ROIs from 2p cell segmentations
Other
0 stars 0 forks source link

Fix blurry video artifacts #94

Closed pickles-bread-and-butter closed 4 years ago

pickles-bread-and-butter commented 4 years ago

Overview

Screen Shot 2020-05-12 at 1.50.22 PM.png It can be seen from the above screen shot that the objects in the frame are blurry. This could be the result of codec issues using constant bit rate over constant quality. To maintain constant quality in a video the bitrate must change throughout, restricting the bit rate leaves highly changing parts of the video without the quality of less changing sections.

Initial options to check out:

Notes

ffmpeg constant quality docs bit rates in videos ffmpeg commands in imageio

Validation Criteria:

Both full-size and 128x128 videos play on modern Chrome and Firefox browsers. The videos load and play consistently, line up with the traces, there are no visible artifacts, and they can be used by labelers to make classification decisions. This should be demonstrated for videos that come from at least five experiments, since artifacts are not present for every video. For testing, create a simple webpage that plays the videos from local files. For final validation, create a SageMaker job and invite the dev team to test.

njmei commented 4 years ago

imageio-ffmpeg looks to literally just stitch together an ffmpeg command. Here is the relevant line for what modifying 'bitrate' does. Which is probably not quite the desired thing for the vp9 encoder.

https://github.com/imageio/imageio-ffmpeg/blob/974e92e699676822971d66765a2787ecac3a0788/imageio_ffmpeg/_io.py#L383

kschelonka commented 4 years ago

Updated title and validation criteria.

njmei commented 4 years ago

Resolved by #100