Can face recognition be enhanced, to pick up faces in the distance, where less reconstruction/processing is required to make someone look human? If someone is too far away, the rest of the image is processed nicely, but the resulting distant faces instantly make the image unusable.
Thanks
The AI models rely on existing image data so if the original lacks detail or if the resolution is too poor, the app will struggle as well. That’s what we see happening here. Try using just the Upscale filter to enhance the resolution of the entire image.
Thanks Tim,
I actually already used the upscale filter to increase the size. I should’ve marked the image up better but basically:
LEFT image: Original unedited.
RIGHT image: Upscaled and then cropped in to show you the result in more detail.
As well as the existing image data, I thought (or rather hoped) that if there was poor image data, the Ai model would refer to it’s ‘image bank’ based on similar scenarios, and be able to ‘build’ the pieces that it can’t collect from the existing image.
For something like this, the face wouldn’t actually need to have a lot of detail, but it would be good if it could give a better approximation of the human face, based on the shapes presented in the image.
Do you think this might be something that will happen in future releases?
1 Like
We’re definitely working on improving our AI models. We’re toying with the idea of being able to train the model with specific faces, eg. you could input high quality pictures of yourself to recover an image where the face might need Face Recovery and it builds it based on your specific images.
This is not currently in the works but we are improving the models in other ways at the moment.
1 Like
This sounds like a nice feature to look forward to, hopefully it goes ahead one day as I’m sure this would make an excellent feature for older photography - train the software on your own face, with multiple images uploaded at different angles so it can build a ‘model’ of certain individuals! It would also be great, as mentioned above, if there could be a feature based around distant faces too, where low detail is required, just to ensure that they are still considered when the algorithm looks at the entire image. Cheers!
1 Like
