Feature Requests

[EddieGriswold]

First off, I want to say thank you for your hard work on a project that seems to do what I can't even find in paid software. I made a small contribution a few weeks ago and while it's not much, I hope it's seen as a token of my appreciation for this. I'd love to see this grow in popularity and contributions from those with the skills and time.

While getting pretty familiar with the app, I've ran into some issues that point out gaps in the tool. If you have any ideas for ways to workaround these with existing features, I'm all ears.

1. Filters for media without any face detections (speed up the process of resolving undetected faces in media)
2. The ability to selectively reprocess specific media (e.g.: after tweaking settings, troubleshooting undetected faces, etc.)
3. The ability to increase the number of previews under the Faces > Suggested view. Being limited to 20 faces per iteration for people that have thousands of photos results in a lot of repetitive paging for matches
4. Troubleshoot issues with detecting faces when persons are laying down or are upside down in the media. It seems the model doesn't match on faces that are sideways or upside down, laying left to right, even with face showing. This is a big pain and results in a lot of manual rotation of media (Note: this can be resolved by rotating the photo to orientate the face upright). I can provide examples of this if needed for training or testing

[EinAeffchen]

Thank you very much for your support!
Regarding your feature requests:

1. Filters for media without any face detections (speed up the process of resolving undetected faces in media)

What exactly do you mean? A filter to find media for which no faces could be detected?

2. The ability to selectively reprocess specific media (e.g.: after tweaking settings, troubleshooting undetected faces, etc.)

If I understand it correctly, you want to select only a few images/videos to post-process again?
I think actively selecting them would be too much work for the usual library in the size of tens of thousands of images, but perhaps an option to reprocess all media where no faces have been detected yet might help?

3. The ability to increase the number of previews under the Faces > Suggested view. Being limited to 20 faces per iteration for people that have thousands of photos results in a lot of repetitive paging for matches

Fair point, it has honestly bothered me sometimes as well, I'll increase the amount by default and add an option into the settings in the next release.

4. Troubleshoot issues with detecting faces when persons are laying down or are upside down in the media. It seems the model doesn't match on faces that are sideways or upside down, laying left to right, even with face showing. This is a big pain and results in a lot of manual rotation of media (Note: this can be resolved by rotating the photo to orientate the face upright). I can provide examples of this if needed for training or testing

That is odd, the rotation and alignment should happen already, as I am using insightface with the buffalo_l model, which computes an affine transformation based on the 5 key points (eyes, nose and mouth corners) and 106 facial landmarks.
But if you can provide example images that fail to be detected, I can try to debug that issue.

Thanks for the feedback!

[EddieGriswold]

1. Filters for media without any face detections (speed up the process of resolving undetected faces in media)

What exactly do you mean? A filter to find media for which no faces could be detected?

Correct. I am working through my library to make sure all faces are detected and assigned to the correct person. In pursuit of 100%, it's very difficult to ensure every photo or video with a face in it has been matched to someone without a filter to know what is left.

3. The ability to selectively reprocess specific media (e.g.: after tweaking settings, troubleshooting undetected faces, etc.)

If I understand it correctly, you want to select only a few images/videos to post-process again? I think actively selecting them would be too much work for the usual library in the size of tens of thousands of images, but perhaps an option to reprocess all media where no faces have been detected yet might help?

I think that would be a sufficient workaround, if that's easier to implement. I've had a few photos and videos that I've had to delete then re-import and process to get picked up after making tweaks to face detection parameters. That would solve for that.

5. The ability to increase the number of previews under the Faces > Suggested view. Being limited to 20 faces per iteration for people that have thousands of photos results in a lot of repetitive paging for matches

Fair point, it has honestly bothered me sometimes as well, I'll increase the amount by default and add an option into the settings in the next release.

Thank you!

7. Troubleshoot issues with detecting faces when persons are laying down or are upside down in the media. It seems the model doesn't match on faces that are sideways or upside down, laying left to right, even with face showing. This is a big pain and results in a lot of manual rotation of media (Note: this can be resolved by rotating the photo to orientate the face upright). I can provide examples of this if needed for training or testing

That is odd, the rotation and alignment should happen already, as I am using insightface with the buffalo_l model, which computes an affine transformation based on the 5 key points (eyes, nose and mouth corners) and 106 facial landmarks. But if you can provide example images that fail to be detected, I can try to debug that issue.

I can do that. Do you have a spam email that I can send those to?

[EinAeffchen]

Correct. I am working through my library to make sure all faces are detected and assigned to the correct person. In pursuit of 100%, it's very difficult to ensure every photo or video with a face in it has been matched to someone without a filter to know what is left.
I added various new filters in the 0.6.4 release and the possibility to run single processors again from within the config -> Processors tab!

I haven't gotten around to enable running them for select files though.

Through the new filters, I found various images where my system did not detect faces either, so I think that should be enough data for me to tweak the detection, otherwise I'll come back to you for your samples!

[EinAeffchen]

I added some new features that should help in finding all faces and assigning them, including filters for select scenes on videos.
You can now also run single processors on select images/videos.

I can do that. Do you have a spam email that I can send those to?
Not really, can you use some image upload service and post the links if they are not too private?