I'd assume you could detect a difference between a completely/nearly completely covered LED shield and one with a large gap between your hand at the top of the shield.
Would you be able to use the LEDs as light sensors and eliminate the switch component entirely? Maybe a tubular shield over each LED so it doesn't interact with others around it then covering it with your finger could be read as toggling the state?
Generating many conversions for all photos incurs both storage and computational overhead that may never pay off. By only generating the required images on the fly for photos we avoid this overhead while allowing more flexibility. If watermarks or attribution change in the future it's easier to have the resize service in place to handle this as images expire from the CDN than to do a batch job to pre-process the updated conversion requirements for all existing photos.
Also note that photos are stored indefinitely, so any additional permanent storage (new conversions) would add additional storage costs, miniscule as they might be, forever.
I'd like to be able to have users upload directly to S3 (obviously less infrastructure and code to maintain) but without being able to provide immediate feedback on the upload I've found it to be preferable to have our own application servers in the upload path. This lets us immediately detect unsupported formats and act on the completion of the upload (whether successful or failure) without delay.
VIPS has been great to work with. The docs are generally very good and the performance has been great. The author (https://github.com/jcupitt/) has always been responsive to issues and questions and seems to be constantly working on the project.
My only, relatively minor quibble, is dev work happening on master, but when it's largely a one-man project who am I to judge. So far we haven't come across any problems that weren't either self-inflicted or already fixed in master.