KichangKim / DeepDanbooru

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

How to output the result to txt? #90

Open Cdanzhi opened 1 year ago

Cdanzhi commented 1 year ago

I deployed the project locally, but after I used ‘deepdanbooru evaluate.......’, I found that the result was just printed in cmd. How to save the results separately as the corresponding txt file for each picture?

Thank you very much!

Jena8765 commented 1 year ago

If you have GIT you can open a git bash instead of a cmd window and run the command deepdanbooru evaluate [image_file_path or folder] --project-path [your_projectfolder] --allow-folder | tee -a output$(date +'%Y-%m-%d_%H-%M-%S').txt and it will output in the bash window and a time stamped text file but it saves every output into a single text file though it does sort within that one text file, it isn't hard to write a code that can take that one text file and split it into many but as far as I know outside of editing DeepDanbooru code there isn't a way to do it directly

kamuridesu commented 1 year ago

If you have GIT you can open a git bash instead of a cmd window and run the command deepdanbooru evaluate [image_file_path or folder] --project-path [your_projectfolder] --allow-folder | tee -a output$(date +'%Y-%m-%d_%H-%M-%S').txt and it will output in the bash window and a time stamped text file but it saves every output into a single text file though it does sort within that one text file, it isn't hard to write a code that can take that one text file and split it into many but as far as I know outside of editing DeepDanbooru code there isn't a way to do it directly

Why??? you can just pass --save_txt to evaluate and it generates a .txt to each file!

soleilvermeil commented 1 year ago

You can indeed add --save-txt (note the dash, not an underscore) to your command, which will create a file with the same name as your image (but with .txt obviously) containing its tags. For the moment only the tags are saved; the confidence of each tag is lost during the process.