MobileNet. Update V3 models and description

Author: dmitrygolovkin

caffe_models/mobilenet/caffe_model/mobilenet_v3/README.md

@@ -0,0 +1,16 @@
+The model origin is: https://github.com/tensorflow/models/tree/master/research/slim/nets/mobilenet  
+Jira record: https://jira.internal.synopsys.com/browse/P10019563-39535  
+
+MobileNet v3 model could be converted from frozen pb or tflite to caffe model.  
+
+Conversion command:
+1. Model converted from TF: pb_converted/v3-small_224_0.75_float_pb_convert.prototxt
+> evconvert tf2ev -g v3-small_224_0.75_float.pb -p input 1 224 224 3
+
+2. (Old) model converted from tflite: v3-small_224_0.75_float_convert.prototxt
+> evconvert tflite2ev -m v3-small_224_0.75_float.tflite
+This model contains Permute layers that should have been optimized away during the conversion.
+The model converted from TF does not have these layers
+
+The *_nohardsiwsh.prototxt models are generated manually by changing all HardSwish layers into ReLU layers
+

caffe_models/mobilenet/caffe_model/mobilenet_v3/pb_converted/v3-small_224_0.75_float_pb_convert.caffemodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5886403309fe4474752368d01824500f5bf5184f50a3288397f17470b1d3d64d
+size 8313163