Hi,
Since I cannot set multiple output layers according to:https://developer.qualcomm.com/forum/qdn-forums/software/snapdragon-neur...
So I consider to concat all the outputs into ONE output, but it seems that concat operation only support operands with the same shape,
for example, I have 3 outputs with shapes (1, 4332), (1, 2000), (1, 500), tensorflow support concat operation on the last dimension, when I convert the model to dlc file, it would report:
/home/damon/work/snpe-sdk/snpe-1.8.0/lib/python/converters/tensorflow/layers/concat.py:80: RuntimeWarning: error_code=1004; error_message=Layer parameters combination is invalid. Layer final_output: input BoxPredictor_1/box_encodings_1:0 has size 2000 along axis 1, should match output dim (4332); error_component=Model Validation; line_no=517; thread_id=139899944376064
descriptor.axis)
The error log indicates that output dimension should be 4332, it asks that all the operands must be shaped to (1, 4332).
Does that mean concat operation supports only operands with the same shape?
Any comments would be appreciated! Thank you.
Dear SNPE SDK developers,
Does concatenation suuports only tensors with the same shape?
Any comments would be appreciated! Thank you.