VGG based Face Embeddings takes a long time

[idata]

I am trying to run this network for face recognition. http://www.robots.ox.ac.uk/~vgg/software/vgg_face/

The inference time is 750ms which seems too high. What can be the issue here?Network Summary Detailed Per Layer Profile Bandwidth time # Name MFLOPs (MB/s) (ms) =============================================================================================================================== 0 conv1_1 173.4 304.2 8.510 1 conv1_2 3699.4 663.1 83.248 2 pool1 3.2 832.7 7.356 3 conv2_1 1849.7 421.7 33.017 4 conv2_2 3699.4 474.0 58.752 5 pool2 1.6 923.1 3.318 6 conv3_1 1849.7 171.7 43.405 7 conv3_2 3699.4 177.9 83.779 8 conv3_3 3699.4 178.0 83.765 9 pool3 0.8 932.3 1.643 10 conv4_1 1849.7 136.6 41.684 11 conv4_2 3699.4 172.4 66.082 12 conv4_3 3699.4 172.6 65.987 13 pool4 0.4 926.7 0.826 14 conv5_1 924.8 317.1 19.627 15 conv5_2 924.8 316.9 19.634 16 conv5_3 924.8 314.6 19.780 17 pool5 0.1 880.9 0.217 18 fc6 205.5 2157.2 90.881 19 fc7 33.6 2138.0 14.971 20 fc8 21.5 2106.5 9.728 21 prob 0.0 10.9 0.458 ------------------------------------------------------------------------------------------------------------------------------- Total inference time 756.67 -------------------------------------------------------------------------------------------------------------------------------

[idata]

@AshwinVijayakumar can you assist here? Why is it taking too long

[idata]

@gauravmoon Hey, how did you make it work. While compiling, I am getting this error:

"mvNCCompile VGG_FACE_deploy.prototxt -w VGG_FACE.caffemodel -s 12"[Error 20] Toolkit Error: Input Layer must be in an input_shape construct.

Can you please guide me with this. Thanks :)

[idata]

@sri_mash Please visit https://movidius.github.io/ncsdk/Caffe.html and take notice that there is a specific input format that we support. The input for the prototxt file should be in this format:

name: "GoogleNet" input: "data" input_shape { dim:1 dim:3 dim:224 dim:224 }

[idata]

@Tome_at_Intel Thanks a lot! :)