Error in Infer

Yatish_M_Intel · ‎07-09-2018

Hello Everyone

I am trying to convert frozen tensoflow model to IR and getting following error

Cannot infer shapes or values for node "detector/yolo-v3/strided_slice".

strided_slice node is created in tensorflow by the line

inputs = inputs[:, 2:-2, 2:-2, :]

What am i missing ?

Thanks

Yatish

Monique_J_Intel · ‎07-09-2018

Hi Yatish,

If the input dimensions are in that format MO cannot interpret them for that node. Can you send the full log of the error from MO?

Kind Regards,

Monique Jones

Yatish_M_Intel · ‎07-09-2018

Hi Monique

I have attached a logfile with the comment. Thanks for looking into it.

Best Regards

Yatish

Yatish_M_Intel · ‎07-09-2018

After modifying my tensorflow model to folowing

test = inputs.shape

# trim back to desired size
inputs = inputs[:, 2:test[1]-2, 2:test[2]-2, :]

i moved forward with the conversion, but it still failed with following error message

[ 2018-07-09 12:49:46,680 ] [ ERROR ] [ main:230 ] -------------------------------------------------
[ 2018-07-09 12:49:46,680 ] [ ERROR ] [ main:231 ] ----------------- INTERNAL ERROR ----------------
[ 2018-07-09 12:49:46,680 ] [ ERROR ] [ main:232 ] Unexpected exception happened.
[ 2018-07-09 12:49:46,680 ] [ ERROR ] [ main:233 ] Please contact Model Optimizer developers and forward the following information:
[ 2018-07-09 12:49:46,680 ] [ ERROR ] [ main:234 ] The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()
[ 2018-07-09 12:49:46,680 ] [ ERROR ] [ main:235 ] Traceback (most recent call last):
File "/opt/intel/computer_vision_sdk_fpga_2018.1.267/deployment_tools/model_optimizer/mo/main.py", line 222, in main
    return driver(argv)
File "/opt/intel/computer_vision_sdk_fpga_2018.1.267/deployment_tools/model_optimizer/mo/main.py", line 190, in driver
    mean_scale_values=mean_scale)
File "/opt/intel/computer_vision_sdk_fpga_2018.1.267/deployment_tools/model_optimizer/mo/pipeline/tf.py", line 182, in tf2nx
    fuse_mul_add_sequence(graph)
File "/opt/intel/computer_vision_sdk_fpga_2018.1.267/deployment_tools/model_optimizer/mo/middle/passes/fusing.py", line 452, in fuse_mul_add_sequence
    is_fused = fuse_linear_sequence(graph, node)
File "/opt/intel/computer_vision_sdk_fpga_2018.1.267/deployment_tools/model_optimizer/mo/middle/passes/fusing.py", line 405, in fuse_linear_sequence
    if any([x != 0 for x in add]) and any([x != 1 for x in mul]):
ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

I have attached logfile with the comment.

Thanks

Yatish