I hate spiders. Like, why do they gotta have so many legs? Eight?? For what?? That’s six more than I’m comfortable with. And they don’t even walk normal—they scuttle. Scuttling is not an acceptable form of transportation. It’s creepy and it needs to be banned from showing in my browser. (I’d like to appreciate some games that provide an arachnophobia filter option such as Grounded and Satisfactory.)
Recently when I was browsing Hugging Face, I remembered that there are a lot of Transformers based image classifier tools which I can utilise to overcome (by overcome I mean block) my fear of spiders.
There ARE existing solutions that blocks spider images when browsing. However, none of them is privacy-friendly. Most of them upload images to a cloud server running closed-source software to do the processing. And the rest of them simply runs a large model directly inside your browser which is slow and memory inefficient.
SwinV2 Based Spider Detection
SwinV2 is a very advanced image detection model that’s supported in Hugging Face’s Transformer framework. What makes this model superior is that it can be run very efficiently with Transformers framework on all major GPU brands including AMD, Intel, and Nvidia. When tested on an AMD RX7900XTX 24GB GPU, image (spider soon) classification can be done in less than 200ms.
Structure
This will be a 2 part software.
Front end: A front-end running in JS as a TamperMonkey Script. It uploads all images larger than a user defineable size to the backend
Back end: Backend is running a http server locally in Python that uses Transformers to process images locally with GPU acceleration. The SwinV2 model might need to be finetuned to accurately detect spiders. But fortunately there are spider datasets available on Hugging Face.
…
Anyway. No shade to Mother Nature, but spiders need a software update. Preferably one that includes a “stay outdoors forever” patch.