Distortion of facial features due to fast-moving subjects or insufficient pixel data
When a subject moves quickly or there is insufficient pixel data, facial features may become distorted. This limits the ability to accurately restore faces during upscaling.
The need to specify and improve the facial features of specific individuals or groups
To enhance the facial features of particular individuals or groups, there should be a feature that allows users to designate these faces. In scenes with multiple characters, it can be difficult to improve each face separately, so it’s necessary to allow users to select specific individuals or groups for enhancement.
Proposed Solutions:
Adaptive Detail Enhancement
When users identify scenes where the face has been distorted, the system will use portions of the track that contain relatively more data to restore the face with greater detail.
Supplemental Data Provision
If there is insufficient data, users can provide images or video clips of the individual, allowing the AI to upscale the face according to the specific environment of the track.
I think Topaz don’t like my suggestion, my post was deleted months ago?! How it is, I suggested two new paramaters you can set:
“face recovery strength” as higher as harder face recovery itself happens.
“face recovery sensitivity” as lower the value is set, as only good recognizable faces gets recovered.
So for the long distance worse recovery disorted faces, you can set “sensitivity” to very low and “strength” on default or set it to high and then you get good face recovery as we known but on faces with low pixel data, there no recovery happens!
I second this suggestion, it is called face sharpen and face soften within Hitpaw Video Enhancer and that software is doing a much better job in low res face restoration than Topaz.
I can tell you it must be due to some technical issues or outdated drivers on your system because for me it works flawlessly.
Of course this is not the forum to discuss HitPaw issues so you may want to check their self-help articles. You can also DM me I may be able to help to figure it out as I work as technical troubleshooting customer support agent myself for a different company.
Yeah good idea, and another enhancement could be to assign resource focus on certain priority individuals for example Beyonce in a dance scene over less needed background extras having too much of that processing ‘wasted’ on them.
Technical Challenges:
Video scenes often involve multiple individuals overlapping and moving in real-time, which makes achieving detailed and natural upscaling for specific individuals or groups extremely challenging.
• While working on general objects might be relatively straightforward, applying this level of detail to human faces or specific individuals is significantly more complex.
Required Technologies:
This isn’t just about utilizing machine learning or AI modeling. It also requires integrating advanced object tracking technologies and ensuring the software operates seamlessly without bugs, further adding to the complexity.
Legal Considerations:
The technology must avoid infringing on patents or copyrights held by other companies. Developing this into a proprietary solution for Topaz is likely to require significant time and effort.
Personal Hope:
Despite these challenges, I genuinely hope to see this technology implemented in Topaz Video someday. This could truly be a game-changer for the platform.
