I took machine learning classes in school and am now trying to create and train a CNN for my work. I did not realize how spoiled I was that my class gave me the thousands of training images all collected and labeled. Eye opening how much work that part is
Like going through google earth and sniping 7000 images for training is no joke. I had never thought about it before cause I’ve only ever had to worry about the code side of things.
9
Anonymous12w
Honestly, the real world and doing things on your own really makes you realize how much more work things are in general rather than in school. School really is just a stepping stone into another 99 levels 😭
8
🏆
Anonymous12w
I can't offer training photos, but I can offer train photos.
5
Anonymous12w
Ahh yes the class that made me see multivariable calculus in the theory in it it all made sense and finally got it to use it in real world
3
Anonymous12w
isn’t that usually a different job responsibility?
1
Anonymous12w
Not horrible if you can script it out. If it’s something a tad more complicated, say medical scans, you kinda are screwed if you don’t got a spider that catalogues from the web…
1
Anonymous#212w
A lot of times 80% of ML engineers jobs is to collect and clean up data.
1
Anonymous#412w
It’s the endings of guard rails on roads. So, not really something I can script out. I’m trying to make a rough draft model using as little hand classified data possible that can help me roughly classify a big drop for the actual model. And then that has been awful
1
Anonymous#212w
Hahaha, not when you work for the government
1
AnonymousOP12w
How big does your dataset need to be? I.e. can you and friends drive in separate directions and collect your data points in person?
1
Anonymous#412w
Well, currently working with 600 photos for 11 classes and the overfitting is insane. The internet says I need atleast 1000 photos per class for a minimum accurate model. So I’m kind of in a pickle