Large Dataset for Machine Learning Project About Climbing

Hi all!

I'm taking a Machine Learning class and our final project is to pick something cool and do it so

I've decided to try and teach a model how to draw a good climbing path along a new rock face.

To do this I need a fairly large dataset of rock faces without a route drawn on top and with a route drawn over it.

Does anyone know where I could find easy access to such files?

Thanks!

EX:
Something like this and a picture without the route drawn on it.

What is a "dataset of rock faces"? Is that a photo of a rock face?

Just edited my post^

Cody:  You say you need " a fairly large dataset of rock faces without a route drawn on top and with a route drawn over it."

What's "fairly large" ....10, 20, 40, 80, 200,  more  ???

Most folk on Mtn project who post a photo with the route drawn onto it have the "raw" photo in their own computer.

[Edit: I misread your post.]

Good luck finding such a dataset (not sarcastic, I do think that would be neat!)

As for dataset size, the larger the better, but ~300 is a good place to start.

You might try contacting some guidebook authors. They may have what you want. This seems like a very difficult project and as Nathan said above, choosing something different would seem like a good idea. The only real rock routes that I think would "make sense" to a computer would be cracks.

ClimbingOn wrote: The only real rock routes that "make sense" are cracks.

[Edited your post too] :p

Nathan Sollenberger wrote:

[Edited your post too] :p

Ha, nice!

Great, thanks for the direction!

As for it being difficult, I don't think it will be too hard if I can get the data.

We have already done a cancer image detection project with the same idea and it gets pretty accurate.
 The goal is just to see if I can get it to work.

Cody Kesler wrote: Great, thanks for the direction!

As for it being difficult, I don't think it will be too hard if I can get the data.

We have already done a cancer image detection project with the same idea and it gets pretty accurate.
 The goal is just to see if I can get it to work.

Hey!
I know it's been a long time but did you end up getting a dataset for that ever?

I wanted to build a similar machine learning project and was wondering if you ever found a dataset for that.

Thanks!

You might try contacting some guidebook authors. They may have what you want.

They would be insane to turn this material over to generative AI (that will replace them).

Tide has turned. It was a neat project back then.

Now, you're stealing topos to feed into an LLM. That's gauche today, try again in another 6 years.

Cherokee Nunes wrote:

They would be insane to turn this material over to generative AI (that will replace them).

Insane would be trusting a topo generated by AI trained on a dataset as described above.

Drawing a useful topo requires real world verification that the claimed route actually exists. No way am I trusting an AI topo generated from a picture that doesn't have enough resolution to even show bolts.  Maybe if someone with actual beta fed it a rough description and verified the output.

I scraped Mountain Project for a grad school project a few years ago. I put the data on Kaggle: https://www.kaggle.com/datasets/pdegner/mountain-project-rotues-and-forums

It doesn't have pictures, but it does have the URLs of the routes which would make it easy to build your own scraper.  

nat han wrote:

Insane would be trusting a topo generated by AI trained on a dataset as described above.

Drawing a useful topo requires real world verification that the claimed route actually exists. No way am I trusting an AI topo generated from a picture that doesn't have enough resolution to even show bolts.  Maybe if someone with actual beta fed it a rough description and verified the output.

Perhaps images from Gigapixel?

Robert Hall wrote: Most folk on Mtn project who post a photo with the route drawn onto it have the "raw" photo in their own computer.

There's a checkbox under the photo to toggle the overlay on and off, so yes you can download both versions.

Example:

https://mountainproject.com/assets/photos/climb/126085416_large_1714629818_topo.jpg

vs

https://mountainproject.com/assets/photos/climb/126085416_large_1714629818.jpg

Heliodor Jalba wrote:

There's a checkbox under the photo to toggle the overlay on and off, so yes you can download both versions.

Example:

https://mountainproject.com/assets/photos/climb/126085416_large_1714629818_topo.jpg

vs

https://mountainproject.com/assets/photos/climb/126085416_large_1714629818.jpg

I'm guessing if he couldn't figure this out on his own he didn't end up doing it. 

Seems dumb anyway? There's cleaning involved when putting up routes so the face would change etc