| Summary: | suggestions to improve face detection workflow | ||
|---|---|---|---|
| Product: | [Applications] digikam | Reporter: | Chris Thieme <thieme_ch> |
| Component: | Faces-Workflow | Assignee: | Digikam Developers <digikam-bugs-null> |
| Status: | REPORTED --- | ||
| Severity: | wishlist | CC: | caulier.gilles, metzpinguin, michael_miller |
| Priority: | NOR | ||
| Version First Reported In: | unspecified | ||
| Target Milestone: | --- | ||
| Platform: | unspecified | ||
| OS: | All | ||
| Latest Commit: | Version Fixed/Implemented In: | ||
| Sentry Crash Report: | |||
|
Description
Chris Thieme
2020-12-11 09:01:20 UTC
This wish is also problematic. Users don't just group JPEG and RAW files, we can't avoid a full scan. Even with a series of images, the image has to be scanned because the position of the face rectangles changes or new ones are added. The information from a previous image does not help at all. In addition, the images are not processed in order in multiprocessor mode. The face recognition run is relatively quick and you can confirm multiple images at once. Maik Michael, This file is about Clustering. Right ? Gilles Hi Gilles, I agree with @Maik that this request is problematic. I think the correct solution is to make face detection and recognition faster for each image individually. Using other images to try to identify faces different images I don't think is a viable solution. I do agree that we should make sure we are grouping images together in the UI based on folder and timestamp. This would make it easier to use shift-click or ctrl-click to select and tag similar faces. Cheers, Mike In my experience, the scanning time for CR2 files was considerably longer than for jpg files. Hence my suggestion to reuse the detected face tags from the jpgs for the raw images. I completely agree that grouping other than raw to jpg will not work. Hi Chris, Yes. Scanning a RAW file, and your case a .CR2 will take longer because we have to convert the RAW to jpeg for face recognition. The issue is that the JPEG image may or may not be exactly the same size as the .CR2. I shoot both Sony and Canon, and the camera jpegs are smaller dimensions than the raw. The face recognition is very sensitive to having the exact pixel locations for the face. We can't easily upscale the jpeg dimensions to the RAW dimensions and be absolutely sure the pixels haven't shifted a little bit. I think you'll find the new face detection system to be much faster than the old one, especially if you were using YOLO. Hope this helps. Cheers, Mike |