Remember when you had to manually label thousands of images for a simple model?

I was looking at some old project files from maybe 2018, and it hit me how much the data prep step has changed. Back then, for a basic image classifier to tell a dog from a cat, I'd spend a solid week just drawing boxes and typing labels. It was brutal, mind-numbing work. Now, with the auto-labeling tools in platforms like Roboflow, you can feed it a few hundred images and it gets you 80% of the way there in an hour. The big shift, in my view, came from better pre-trained models that can suggest labels, not just from faster computers. It turned the most boring part of the job into something you can actually get through in a morning. Has anyone else gone back to an old project and been stunned by how much time a new tool would save now?

3 comments

3 Comments

betty_perry2410d ago

The progress in just a few short years is honestly staggering.

lucast8110d ago

Actually, @betty_perry24, the core tech for this was pretty solid over a decade ago. The real speed-up has just been in the last five years.

anthonynelson10d ago

Ever notice how this keeps happening? Like @lucast81 said, the tech was there, but now it's just baked into everything. I see it with photo apps that sort my pictures or even my email spam filter getting way better without me doing anything.