[fuzz_junket]

Does anyone know what this actually does? There's no explanation in the documentation, and the only screenshot is of the build process.

I get the impression whatever it is requires metadata from a digital camera, which isn't present in my photos because I like to shoot film and scan it. My first thought was, "Wow, how is it going to analyse the subject matter and classify all my scans in 'seceonds'?" My second thought was, "Ah, it won't." The author is clear that this tool is based solely on their own workflows, which is fair enough, but I'd at least like to know what those workflows are.

[rlupi]

It moves photos between local files, dropbox, and google drive, depending on how you configure it.

From a quick look at the code, so far you can achieve the same with exiftool if you mount those sources as local drives. There isn't much else yet, but it wouldn't take a lot of work to hook this up with a vision models to add some labels or metadata.

For example, I normally run the following to import files from an SDCard to a local folder and organize them by date:

exiftool -r -d ~/Pictures/%Y-%m/ '-FileName<DateTimeOriginal' -o . /Volumes/LEICA\ M/DCIM/100LEICA

A bit off-topic: I have the same problem that you face with shooting film with Leica and manual lens.

It would be interesting to see if there are tools to automatically populate metadata based on the content:

1) field of view + rough estimation of lens data can be computable directly from perspective cues

2) this can be narrowed down to specific objectives either:

  2a) using a short list (all objectives a specific photographer owns), or 

  2b) using enough lens data + images (e.g. something that DXO or Adobe could do)

While I am skeptical about doing (2b), it should be possible to do (2a). A middle ground is to manually label a few images and use semi-supervised learning to propagate them to the rest of one's photo collection.

[fuzz_junket]

> From a quick look at the code, so far you can achieve the same with exiftool if you mount those sources as local drives. There isn't much else yet, but it wouldn't take a lot of work to hook this up with a vision models to add some labels or metadata.

Thanks for walking me through it, and for the handy example with exiftool. I like that you're doing this with command-line tools because you can see what's going on more easily. I think it's more accessible than requiring a Java + Gradle build environment.

> 1) field of view + rough estimation of lens data can be computable directly from perspective cues

> 2) this can be narrowed down to specific objectives

Really! That is so cool, I wasn't aware that kind of thing was possible. I never write down the lenses I've used (because lazy) so extracting that kind of metadata from the image would be an interesting exercise.