Skip to content

Conversation

@Breakthrough
Copy link
Owner

@Breakthrough Breakthrough commented Feb 22, 2025

Adds a new performance hint property to detectors to indicate if they have a significant performance benefit when using a downscaled version of the frame with minimal accuracy loss. #401

Disable downscaling if set to auto for HashDetector (detect-hash) as it internally resizes the image to a specific size already.

@Breakthrough Breakthrough added this to the 0.6.6 milestone Feb 22, 2025
@Breakthrough Breakthrough marked this pull request as draft February 22, 2025 04:30
@Breakthrough
Copy link
Owner Author

Breakthrough commented Feb 22, 2025

This actually seems to make things slower, and while it does have marginal improvements on BBC, is less accurate on the AutoShot database. Will just leave this as-is.

Overall Results for detect-hash on dataset AutoShot:
Recall: 56.70, Precision: 75.96, F1: 64.93 Elapsed time: 3.23

Overall Results for detect-hash on dataset BBC:
Recall: 92.36, Precision: 75.56, F1: 83.12 Elapsed time: 18.14

@Breakthrough Breakthrough linked an issue Feb 22, 2025 that may be closed by this pull request
@Breakthrough Breakthrough deleted the issue-401-downscaling branch October 2, 2025 03:52
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Remove downscaling for perceptual hash

2 participants