So I’m working on a project where my dataset is just people’s images, and I want to finetune it. I need to use a pretrained net which does face recognition (imagenet’s dataset is not similar to my data).
I’ve ported the popular pretrained tensorflow models from the davidsandberg/facenet Github repo into pretrained pytorch implementations. Pretrained weights are downloaded and loaded into the module on model instantiation, in a manner similar to the torchvision pretrained models. Hope you find it useful!