Thrasher cover similarities

[tkp]

Been experimenting with computer vision models. They do a better job of matching the overall photo content.

Loop di doo


Got rails?


Different tricks, similar upper body


Skater filling the frame


Photo issues


Upper decker


skart


Big sets


Kickflip wallride

[tkp]

the updated version that uses vision transformers is live over at https://shoplurker.com/labs/img-compare/

fullipes
https://shoplurker.com/labs/img-compare/match?model=vt&cover_date=1985-05-01&filename=May1985.jpg&magazine=thrasher

i've also added the new yorker and artforum

bikes:
https://shoplurker.com/labs/img-compare/match?model=vt&cover_date=2025-06-02&filename=6830a4291fd8066f3414a8f5.jpg&magazine=newyorker

glowing rays / orbs
https://shoplurker.com/labs/img-compare/match?model=vt&cover_date=1976-05-01&filename=1976-05-01.jpg&magazine=artforum

[WarmUpZone]

I haven't been SLAPping much lately so I missed this whole thing until now.

Very, very cool and entertaining.
Maybe some of the Thrasher team could use this tool in conjunction with sales figures and find the winningest visual formula. Cross reference that with 4PLYs sorting by skater or trick there is no stopping them.

[Where Am I]

Been experimenting with computer vision models. They do a better job of matching the overall photo content.

That's so sick! Do you know of any good tutorials on how to get started computer vision models?

[Meekin]

This is neat

[tkp]

Expand Quote

Been experimenting with computer vision models. They do a better job of matching the overall photo content.

[close]

That's so sick! Do you know of any good tutorials on how to get started computer vision models?

@Where Am I I started by learning how to generate image hashes using https://pypi.org/project/ImageHash/ (some good insight there). Then I wrote those hashes to a MySQL database and used a SQL query to calculate the hamming distance between them.

DinoV2 and CLIP has a similar process (generate model encodings, write them to a database, then run a vector similarity on them ). This is a good resource https://medium.com/aimonks/clip-vs-dinov2-in-image-similarity-6fa5aa7ed8c6

[Trilogy]

You can add this Carlos Ribeiro switch BS noseblunt slide cover to the first two, different colours but similar angle.