Web application for capturing image scores. Different experiments for capturing the image ratings.
- Score an image
- Rate an image
- Preference
- Similarity
Score an image based on 1 to 10. Skip means give no rating, and back to the last rating to re-rate it.
Get a predicted score, and modify it to the rating you think it should be. --
down by 1.0, -
down by 0.1, +
up by 0.1 ++
up by 1.0.
Pick a image you prefer out of the list of 4 images. Alternative of best of 2 images.
git clone https://github.com/rockerBOO/image_scorer
cd image_scorer
We use images.json
to create a list of images to rate. Images must be stored in images/
in the main directory. (Not ideal, but how it works)
python make_images_json.py images
Puts the images.json
into priv/images.json
These images will be viewable though images/
URL.
Runs the web service for the UI
gleam run
Compiled in 0.01s
Running image_scorer.main
Listening on http://localhost:3030
Hello from image_scorer!
Then you can go to the web service:
http://localhost:3000/
Running the aesthetic predictive and similarity models. Using poetry to do package management.
poetry run uvicorn ae_scorer_server:app --port 3031
Loading CLIP ViT-L-14...
INFO: Started server process [2399229]
INFO: Waiting for application startup.
INFO: Application startup complete.
INFO: Uvicorn running on http://127.0.0.1:3031 (Press CTRL+C to quit)
No cross domain implementation, currently.
gleam test
Not open for improvements as it's still a work in progress, but any feedback is open and welcome.
- Finish rate.html plus/minus
- Finish similarity file upload
- Finish similarity dataset
- Cleanup JS
- Cleanup styling
- Complete model hosting
- Deploy
- Save each latent and CLIP image embeddings for the images
There are no bugs. … Gotcha.
- Preference when one of the items is 404 causes it the crash and can't proceed