Instagram vs. Snap computer vision
Does anyone else feel as if Instagram is *still* 1-2 years behind Snap in terms of advanced face filters? Most face filters on Instagram (dog ears, etc) feel basic like 2016 Snapchat. And the recent gender swap Snap filters feel far ahead of anything IG will ship soon. How is it possible that a bankrupt company in LA has better deep learning models than a Bay Area tech giant like FB?
Yeah I think IG has more important things to worry about than improving face filters tbh
Camera is a pretty important part of IG, no? Why do they build superzoom, boomerang, filters, etc? State of the art camera filters means better engagement. After all, filters is how IG got it’s start. And as a cherry on top it would quicken Snap’s death.
'State of the art camera filters means better engagement' - You'd be surprised to learn that this is not always the case when it comes to engagement numbers. The juice is not always worth the squeeze. Furthermore Instagram leans towards a more restrained aesthetic in it's filters... This is intentional and on brand... Snapchat is a little more 'National Inquirer' while Instagram is a little more 'Conde Naste' or 'Dwell'. But yes... Snapchat is on the cutting edge of computational photography and the possibilities of mobile ar if that is your yardstick. They will be acquired.
It's just a matter of time before IG ships in the same feature !
Want to innovate (try, experiment, oftentimes fail) - come join us. What to copy (success is almost guaranteed) - go join IG. P.S. Both companies innovate IMO though they focus on different spaces.
I wonder is OP currently working in Instagram or snap computer vision now?
I think IG just doesn't care, because even the gender swap filter is not all that difficult to do from a DL perspective.
I am not a DL expert so correct me if I am wrong. Perhaps performing a basic image-to-image gender swap in PyTorch that looks decent on some faces might be simple. Doing it at scale to work on most faces, live rendering on iOS/Android must be non-trivial, no?
I haven't been following computer vision for the last 2 years but no, IIRC pretty good mobile platform frameworks were already coming out and gender swap should be a pretty lightweight algorithm. It helps that the image resolution is not very high, they probably render it in something even lower than the screen resolution and antialias or something.