Did you tested the performance of the model? Such as the time cost of the predict step?

KleinYuan / Caffe2-iOS

Caffe2 on iOS Real-time Demo. Test with Your Own Model and Photos.

MIT License

227 stars 45 forks source link

Did you tested the performance of the model? Such as the time cost of the predict step? #8

Closed raininglixinyu closed 7 years ago

KleinYuan commented 7 years ago

@raininglixinyu

So, in the master branch, we use squeezeNet as an example, which is a compressed small model. If your question is asking the specific performance of this model, please check the gif below:

[X] iPhone 7 plus
[x] DispatchTime Measurement
[X] Elapse of function classifier

squeezenet

Also, in the meanwhile, I am trying to add more Caffe2 models and will have a performance chart in readme soon.

raininglixinyu commented 7 years ago

@KleinYuan Really thanks for your detailed explanation and it helps a lot. The squeezeNet has two versions(v1.0 & v1.1), which one has been used?

KleinYuan commented 7 years ago

@raininglixinyu The two .pb files are original from here. So, squeezeNet (v1.0 && v1.1) are .caffemodel (which works for Caffe) and those two .pb files are for Caffe2 model. It's a good question and TBH, I don't know which specific version they used.

Let me raise this and link in here.

raininglixinyu commented 7 years ago

@KleinYuan Thanks!This issue can be closed.

KleinYuan commented 7 years ago

@raininglixinyu check here