Link to Collages of Watch Advertisements

Danger Will Robinson, danger! I just can't get over how awesome some of these are and what labor of love this is/has been from the OP at WUS.

Great! 👍🏼 Can’t imagine how long this has taken!!!

See below link for full text search and browsing complete with machine translated texts to English for the posted collages:

URL: https://xelrix123.github.io/watch/search.html

There are roughly 5000 individual ads in the dataset posted to WUS. This is incredible collection which screams for an ability to search into it. It took ~ 48h to tun Qwen3.5 vision model against this dataset to OCR the texts, do object identification, and machine translation to English. And then a bit more use of Claude and Codex to create/update the web frontend that allows search and browse to the images with links back to the collages at Google Drive. I will upload later to Github (or here if there is interest) source and prompts used.

Two things I should add: 1) if original author finds the search above not to his liking, I will take it down (no permissions were asked). And 2) I don't want to end up on a hook for maintaining the search 'app' - so consider this abandonware. For now though (as I am pretty excited of these ads and this new capability), I am all ears for feedback on how to improve the search/frontend.

This work builds on some experimentation I had done last year against scans of my meager 'collection' of (physical) watch ads.

Enjoy!

I suppose that would be good sportsmanship to do so. I approached this a bit from the direction of asking for forgiveness later. Consider it done. edit: contacted the OP @ WUS. 🤞

Below shows llm generated summary for the ad (turns out to be quite useful for searching ads with little to no text to translate), and how the original text has been extracted from ad, and then translated to English while doing object detection. Toggle switch on the 'Advertisement Text' line can be used to see the original text (caveat emptor - can't really comment on quality of most translations, but I think they certainly pass on the intent). The buttons on the header take you to original collage in gdrive, and show the raw JSON that the language model outputted for the add.

I personally made the collages myself (and I'm continuing to make some more as I find ads). Thank you all so much for the compliments. Valerio from Italy...🫶

If you have any questions, please ask. I'm happy to share my little knowledge as far as it goes...

...and bear with me if my English isn't so good...

Thanks, if you're referring to those leather straps visible in the "omega 27.jpg" collage, I have one that I use with a Minerva pocket chronograph, just to give an example...

I'm the author and i have no objections. The 914 collages are for sharing with all of you 👍 From google drive the size of the entire collection is about 1,1/1,2 gigabytes and you can download everything you want 👍

These days, in my spare time, I've already reached 940 and found a new resource with several interesting things I didn't have. I'll probably reach 1000 within 2-3 months at most... 💪

Thanks again to gatorcpa for the repost, and I'm glad you liked them. I especially encourage you to look at those related to the supply chain (ancillary) industry (directory "indotto"). We often don't think about how many companies and people are behind the creation of watches.👍

Collages are in 2100x2970 px format if someone wonna print something in A4 page format 😉