search

forresti / SqueezeNet

SqueezeNet: AlexNet-level accuracy with 50x fewer parameters
BSD 2-Clause "Simplified" License
2.17k stars 723 forks source link

deploy.prototxt #1

Closed kaishijeng closed 8 years ago

kaishijeng commented 8 years ago

Do you have a plan to release deploy.prototxt? I plan to run classify.py from Caffe to test squeezenet performance for images I have and compare it with alexnet

.

forresti commented 8 years ago

I would accept a Pull Request of deploy.prototxt if you'd like to create one. It's pretty straightforward...

  1. Replace the two data layers with input_shape{dim: ...}
  2. Change SoftmaxWithProb to Softmax
  3. Remove the accuracy and accuracy_top5 layers

With these three changes, I think you'll have a file that matches the syntax of the bvlc_alexnet and bvlc_googlenet deploy.prototxt files.

kaishijeng commented 8 years ago

Thanks for the instruction. Let me try first and if it works properly, I can submit it for a pull request

On Thu, Feb 25, 2016 at 8:53 PM, Forrest Iandola notifications@github.com wrote:

I would accept a Pull Request of deploy.prototxt if you'd like to create one. It's pretty straightforward...

  1. Replace the two data layers with input_shape{dim: ...}
  2. Change SoftmaxWithProb to Softmax
  3. Remove the accuracy and accuracy_top5 layers

With these three changes, I think you'll have a file that matches the syntax of the bvlc_alexnet and bvlc_googlenet deploy.prototxt files.

— Reply to this email directly or view it on GitHub https://github.com/DeepScale/SqueezeNet/issues/1#issuecomment-189114976.

kaishijeng commented 8 years ago

I create a deploy.prototxt and it runs OK with Caffe. I also notice someone already creates a PR for deploy.prototxt which is the same as my version. I will not submit a PR and close this issue now