Open MariasStory opened 5 years ago
Hi,
Any more information on this, did you actually try it out? On first sight, the model doesn't look very web friendly to me, the raw weights seem to be about 250MB, plus one would have to investigate how a resnet100 architecture performs on the web.
The MobileFaceNet model looks quite feasible.
Will play around with this some time, thanks for sharing!
I downloaded it, and find out it include 3 files: log, .json .params, how can I use it as a Layer in my Keras neural network
The original model is for mxnet. There’s a port to tensorflow/Keras at https://github.com/shaoanlu/face_toolbox_keras
Maybe something to try: https://github.com/deepinsight/insightface/tree/master/detection/RetinaFace Interesting comment: Third-party Models yangfly: RetinaFace-MobileNet0.25 (baidu cloud:nzof). WiderFace validation mAP: Hard 82.5. (model size: 1.68Mb)
The original model is for mxnet. There’s a port to tensorflow/Keras at https://github.com/shaoanlu/face_toolbox_keras
Thanks for your help.
Hi,
Please check/adopt the following model: https://github.com/deepinsight/insightface/wiki/Model-Zoo In praticular LResNet100E-IR,ArcFace@ms1m-refine-v2 looks good.
Thanks