"IndexError: list index out of range" with ncsdk 2.08

idata · ‎11-07-2018

I installed sdk 2.08 on desktop and tried to compile model from the example https://github.com/ihere1/movidius-face. Command line was "mvNCCompile -in "pnet/input" -on "pnet/output" -o p2838.graph p2838/model.meta". I got the following error

  File "/usr/local/bin/mvNCCompile", line 206, in <module>
    create_graph(args.network, args.image, args.inputnode, args.outputnode, args.outfile, args.nshaves, args.inputsize, args.weights, args.explicit_concat, args.ma2480, args.scheduler, args.new_parser, args.cpp, args)
  File "/usr/local/bin/mvNCCompile", line 185, in create_graph
    load_ret = load_network(args, parser, myriad_config)
  File "/usr/local/bin/ncsdk/Controllers/Scheduler.py", line 146, in load_network
    parse_ret = parse_tensor(arguments, myriad_conf)
  File "/usr/local/bin/ncsdk/Controllers/TensorFlowParser.py", line 1168, in parse_tensor
    xyz = (1, 1, int(input_shape[1]))
  File "/home/ales/.local/lib/python3.6/site-packages/tensorflow/python/framework/tensor_shape.py", line 616, in __getitem__
    return self._dims[key]
IndexError: list index out of range

I've found that following code in TensorFlowParser.py cause crash.

top = [get_input(strip_tensor_id(node.inputs[0].name))[0],
get_input(strip_tensor_id(node.inputs[1].name))[0]]
input_shape = node.inputs[0].get_shape()
outputs = node.outputs[0].get_shape()
if len(input_shape) == 4:
xyz = (int(input_shape[2]),
int(input_shape[1]),
int(input_shape[3]))
else:
xyz = (1, 1, int(input_shape[1]))

where inputs are

39 Mul pnet/PReLU1/Mul_2
IN: pnet/PReLU1/alpha/read:0
IN: pnet/PReLU1/Mul_1:0
OUT: pnet/PReLU1/Mul_2:0

and the first input (IN: pnet/PReLU1/alpha/read:0) has the shape "(10,)". At the same time compiler from ncsdk 1 compiled the above network successfully. Is the case possible to fix.

idata · ‎11-13-2018

Hi @Ales did u get any solution for this…..I'm also stuck at this..any help would be highly appreciated…..thanks.