false positives
There were some false positives when classifying round astronomical objects/bodies like planets. I'll add a CC0 astronomy dataset and finetune the model again this July, 2023.
This is solved in v3, after making improvements to the training algorithms, the architecture, and dataset. See a demo at https://huggingface.co/spaces/aivance/safesearch-v3
Great work!
Is the v3-model already available?
Best wishes!
Not yet. I'll release v3 in July/August.
In case you are wondering why aivance has access, aivance is my account 😅
Super cool model!
Basically, what I did for the last month was to thoroughly test 6+ types of edge cases (e.g. graphs, networks, cartoon/animated characters, out-of-frame videos, and a few more). I adjusted the dataset and model training accordingly.
v3.1 is out! Check it out here: https://huggingface.co/aivance/safesearch-v3.1
Cheers! 🤗
The new model is less than 20 MB on disk. Can you believe it? 🤯