The Python Telegram Bot framework connects to Telegram using the RainRatBot login, waits for a specific action (a photo), downloads the photo, and replies with the result.
The Clarifai custom model has two parts, it has a human-readable web interface that can be used to maintain and train the model.
Then there is the Clarifai API (Application Programming Interface), that RainRatBot can upload a photo, and receive a reply in a standard format that it can use.
The training of the model is based on correctly labelled images. Currently although I'm accepting feedback and photos for new submissions, it isn't directly crowdsourced. I look over the feedback and new photos, and upload new training data somewhere within a day to a couple weeks.
I also look over the submissions. If the submission is a fursuit I can identify, and I believe adding it to the training set would improve the model, I will add it to the training set.
- I look at the submissions.
- I do not log who sends the submissions.
- Do not use it if you do not want a human looking at the photo at some point, even anonymously.
- It might be possible to break the anonymity, forensically. I haven't looked into it, but don't rely on absolute anonymity.
No comments:
Post a Comment