I'm trying to get Yolo-v2 to work by using the yad2k project. I want to re-train with my own data, and then convert to NCS.
The conversion from keras to a frozen graph works with:
python3 ../freeze_graph.py --input_graph=NN.pb \
--input_binary=true \
--input_checkpoint=NN.ckpt \
--output_graph=NN_frozen.pb \
--output_node_name='conv2d_23/BiasAdd';
Then in the docker instance where the NCSDK is installed I run mvNCCompile:
mvNCCompile -s 12 NN_frozen.pb -in=Input -on='conv2d_23/BiasAdd'
mvNCCompile v02.00, Copyright @ Intel Corporation 2017
shape: [1, 608, 608, 3]
[Error 5] Toolkit Error: Stage Details Not Supported: FusedBatchNorm inputs mean and variance are not defined. The graph is not created for inference.
Here are the graph files:
https://drive.google.com/drive/folders/1nO35PL6xxt1Z_Ti9gUltSDoJSHkwsLCM?usp=sharing
