Tuxa

8 ms·

1. 1.▲
   by Tuxa 8y ago

   We are not an image hoster after all. Too many problems would be coming along also.
2. 2.▲
   by Tuxa 8y ago

   I cheated a little bit by creating an AI that allowed me to do it faster in a semi-automatic fashion. However, the first batch to train this "labeling"-network was generated manually, which took forever.
3. 3.▲
   by Tuxa 8y ago

   Yup. Multiple u-nets, actually.
4. 4.▲
   by Tuxa 8y ago

   I tried this during development. It works amazingly well but there are also a lot of downsides: the long processing time is just a minor example.
5. 5.▲
   by Tuxa 8y ago

   yup
6. 6.▲
   by Tuxa 8y ago

   agreed. did not expect this interest. should work again.
7. 7.▲
   by Tuxa 8y ago

   we are about to add new servers to handle the load
8. 8.▲
   by Tuxa 8y ago

   What images did you use? Could you give me an example so I can look into this?
9. 9.▲
   by Tuxa 8y ago

   Time, primarily
10. 10.▲
    by Tuxa 8y ago

    It was primarily trained on portrait shots. So not many samples with groups of people. May change in the future.