search

KichangKim / DeepDanbooru

AI based multi-label girl image classification system, implemented by using TensorFlow.
MIT License
2.65k stars 260 forks source link

Clarification of README.md #103

Closed ppbrown closed 4 months ago

ppbrown commented 7 months ago

Could you please add a bit more details in the readme, with specific directory examples?

It talks about "prepare dataset", but does not make it clear where it should live in relation to the "project"?

If I'm understanding things, maybe it would be clearer if you moved step 1, "prepare dataset", to step 6.

Then step 7 becomes

"deepdanbooru evaluate [your-dataset-directory] ...."

KichangKim commented 7 months ago

Hi. Dataset (including sqlite database file) can be located anywhere, you can specify its path in project.json.

Also, if you use pre-trained project (can download it from Releases section) and for just evaluate images, you do not need to prepare your own dataset. It only be needed for training.

ppbrown commented 7 months ago

Maybe I'm misunderstanding use here. Its probably best if I just say want I want to use this for. I have a set of 1,000 images. I'm hoping to use this program to autogenerate .txt tag files for each file. So, definitely using my own data set :)

It might be best to have a streamlined guide that handled this case specifically?

KichangKim commented 7 months ago

Ah, it is simple. deepdanbooru evaluate [your_images_folder] --project-path [project_folder] --allow-folder --save-txt For details, see #90

ppbrown commented 7 months ago

Hmmm. for the "download-tags" part.. I made a free api key, but when I try to use it with your script...

Seems like it pulls an empty file.

json.decoder.JSONDecodeError: Expecting value: 1line 1 column 1 (char 0)

KichangKim commented 7 months ago

If you only want to evaluate, you can skip 1~6. It only need for training. Pre-trained project already has tags.txt.

KichangKim commented 4 months ago

download-tags issue is fixed.