Open MariasStory opened 5 years ago
I must have been doing something wrong, because decreasing the value of "Min Confidence:" did not do anything.
Hmm, could be a bug in the demo.
It would be also nice if the selection on the right bottom detection would be more precise.
Agree, would be interesting to see, if we can train a better model than the current SSD Mobilenet V1, which is still feasible for web apps. In fact I started with training own face detectors, but have stopped working on it. Will pick up on this some time soon again, so thanks for all the links you shared!
As far as the tiny faces in the background are concerned, they will be useless for further face classification tasks however, as they are simply too small in size as an input for any classification network. So I am not totally convinced yet, how such a face detector would benefit face-api,js or web applications in general.
Hi Vincent (@justadudewhohacks), What about a simple test/benchmark with: https://www.windowscentral.com/sites/wpcentral.com/files/styles/xlarge/public/field/image/2014/11/worlds-largest-selfie.jpg
My result using github version:
I must have been doing something wrong, because decreasing the value of "Min Confidence:" did not do anything.
It would be also nice if the selection on the right bottom detection would be more precise.
Reference:
Greetings, Tolik