I want to train a detection model with the “vgg16” and the “mobilenet2 lite” architectures. Do I need to download another base model (.pth file)?

[100375195]

I am working with the jetson-inference repository.

I have the base model “mobilenet-v1-ssd-mp-0_675.pth”. But if I want to train, for example, with network=mb2-lite, do I need another base model? If so, where can I download base models like the mobilenet-v1-ssd-mp-0_675.pth?

Thanks in advance.

[100375195]

PD: Because when I try to train with mobilenet v2 architecture, using as base model mobilenet-v1-ssd-mp-0_675.pth, I get the error “unexpected keys in state_dict”.

[dusty_nv]

Hi @100375195, I’ve only trained this with the ssd-mobilenet-v1, you might want to check the upstream repo for other base models:

https://github.com/qfgaohao/pytorch-ssd

[100375195]

Thanks very much!