Torch Architecture in Pytorch

lee123456 · April 19, 2017, 6:24pm

Hi,

Is it possible to implement the following architecture in pytorch?

cmusatyalab/openface/blob/master/models/openface/nn4.def.lua

-- Model: nn4.def.lua
-- Description: Implementation of NN4 from the FaceNet paper.
-- Input size: 3x96x96
-- Number of Parameters from net:getParameters() with embSize=128: 6959088
-- Components: Mostly `nn`
-- Devices: CPU and CUDA
--
-- Brandon Amos <http://bamos.github.io>
-- 2015-09-18
--
-- Copyright 2015-2016 Carnegie Mellon University
--
-- Licensed under the Apache License, Version 2.0 (the "License");
-- you may not use this file except in compliance with the License.
-- You may obtain a copy of the License at
--
--     http://www.apache.org/licenses/LICENSE-2.0
--
-- Unless required by applicable law or agreed to in writing, software
-- distributed under the License is distributed on an "AS IS" BASIS,

This file has been truncated. show original

I couldn’t find SpatialConvolutionMM,SpatialCrossMapLRN and nn.Normalize

Thanks

lee123456 · April 21, 2017, 8:32am

Any help here?
Just point me to the right docs

smth · April 27, 2017, 12:01am

CrossMapLRN is missing and isn’t wrapped yet. ConvolutionMM is Conv2d and nn.Normalize is simply just implementable using torch.* operations.

Yifan · May 4, 2017, 10:56pm

Hi Smth, thanks for your reply.
But OpenFace has inception layer, any such layer in Pytorch?

Yifan · May 4, 2017, 10:57pm

Hi Lee,
Have you worked this out? I meet the same situation these days, maybe we can discuss how to implement this in pytorch.

smth · May 5, 2017, 2:06am

yifan, google for inception pytorch, you will see some implementations.