36g Rain Tagger is an automated image tagging software.
With it, you can:
- Run a python api tagger that crawls a file system, mapping a set of 10.8k tags to each image.
- Host a web UI to search tagged images. Searching can be done via tags, or image upload.
36g Rain Tagger uses timm and leverages the model SmilingWolf/wd-swinv2-tagger-v3.
It should run on Linux and Windows.
It is named after 36g.
On a fresh install of Linux, you'll need to make sure to have pip and venv installed:
sudo apt install python3-pip python3.12-venvApp install:
git clone https://github.com/fire-eggs/36g-Rain-Tagger
cd 36g-Rain-Tagger
python3.12 -m venv venv
source venv/bin/activate
python3.12 -m pip install -r requirements.txt
# copy configs_copy.toml to configs.toml
# set variables in your configs.toml
cd src/Note: The tagger will automatically download the image tagging model and save it to ~/.cache/huggingface/hub.
The web ui is run with python3.12 web.py and the tagger is run with python3.12 tagger.py.
Original numbers from skwzrd (wd-swinv2-tagger-v3 model):
| Device | Images | Total Time (s) | Time per Image (s) | Model |
|---|---|---|---|---|
| 4060 TI 16GB GPU | 45 | 2.172 | 0.048 | |
| 5700X x 8 CPU | 45 | 21.277 | 0.473 | |
| i7 8665U x 8 CPU | 45 | 76.273 | 1.695 |
I don't have a GPU, but as any timm-compatible model can be used, I've used different models. My system is an i9-13900H x 20 CPU.
| Images | Total Time (s) | Time per Image (s) | Model | Tag-Image Pairs (1) | Found General Tags (2) |
|---|---|---|---|---|---|
| 36,838 | 32,687 | 0.887 | wd-swinv2-tagger-v3 | 854,229 | 6,076 |
| 36,838 | 36,842 | 0.503 | wd-vit-tagger-v3 | 863,254 | 6,209 |
| 36,838 | 65,817 | 1.786 | wd-eva02-large-tagger-v3 | 1,314,034 | 7,221 |
- Tag-Image Pairs: the total number of image/tag pairings generated
- Found General Tags: the total number of "general" tags applied by the tagger for at least one image
0.1s - 0.4s results on hundreds of thousands of images.
Searched 238,302 images in 0.313s and found 25 results.
This is my clone of skwzrd's original project. Kudos for a fun, educational project!