---
license: other
tags:
- pexel
---

# Note on license:
These photos were pulled from pexels. See https://www.pexels.com/license/

Sumary as I interpret it is:
* Dont resell them as-is.
* If you modify them, then selling IS okay
* Don't make any of the idenfiable people in these photos look bad or use for defamatory purposes, etc.
  

More specificaly, they are pulled from the list of urls at
https://github.com/janpf/self-supervised-multi-task-aesthetic-pretraining/blob/main/dataset/urls.txt

which is why "janpf" is in the dataset name.

# Dataset size
Total expanded size for this dataset is 197G

# Image grouping and tagging
For your convenience, images are broken up into archives of under 1000.
Total number of images is over 130,000

They come with a pre-generated .txt for each of them, autogenerated by the WD14 model provided by OneTrainer.
So, tags for them "exist"... but they arent neccessarily HIGH QUALITY tags. Use them at your own peril.

I am starting with only a few sets, but will eventually upload the whole set, I hope.

# Image format
Note that the image sizes are not standardized; they are raw downloads from pexels. 
Some are 3000 pixels to a side. Some are 4000. Some are 6000 !

Also they are a mix of .jpg, .jpeg, and .png

# Tip on isolating images

If you want to find all images with a particular thing tagged, for example, "dog",
then in Linux, you can do something like

find . -name '*.txt' | xargs grep -l dog >/tmp/dog.list


# Acknowlegements
Thank you to the good folks of Pexels.com who have put together a great free resource.
It's still the best place to go if you want a nice human-friendly interface to browse and possibly download
real-world photos of the higest quality (for free!!)

# Samples


![image/png](https://cdn-uploads.huggingface.co/production/uploads/655bca86bb95edf97882ae7c/m1V8WuTKVaGc7EmY0AfLD.png)

![image/png](https://cdn-uploads.huggingface.co/production/uploads/655bca86bb95edf97882ae7c/gLBtg0AD6WEHzth95_vpZ.png)