frame

Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Sign In

Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

FRUSTRUM POINT NET IN NCS

Hello,
I am facing several issues with my architecture of point net. Which I wish to put on the NCS. Please have a look at the architecture.
The get model function defines the model

https://github.com/charlesq34/frustum-pointnets/blob/master/sunrgbd/sunrgbd_detection/frustum_pointnets_v1_sunrgbd.py

and the tf_util files
https://github.com/charlesq34/frustum-pointnets/blob/master/models/tf_util.py

issues are with max pool function and conv2d . Moreover I am not sure if ncs supports concat , slice etc. how to proceed with it.

or also may consider having a look at the following problem.
input was a verctor of shape 1,2048,6,1 and kernel was of size 1,6,1,64
print(output_size) ---> output [Dimension(1), 2048, 6, Dimension(64)]
print(node.outputs[0]) ----> output Tensor("conv1/Conv2D:0", shape=(1, 2048, 1, 64), dtype=float32)
node.outputs[0].set_shape(output_size) ----> in this line of code inside /usr/local/bin/ncsdk/Controllers/TensorFlowParser.py line 370
raise ValueError("Shapes %s and %s are not compatible" % (self, other))
ValueError: Shapes (1, 2048, 1, 64) and (1, 2048, 6, 64) are not compatible

and also in maxpool

where input shape :
(1, 2048, 1, 1024)

and please note I had a kernel size of 2048,1 for max pooling

File "/usr/local/bin/mvNCCompile", line 118, in
create_graph(args.network, args.inputnode, args.outputnode, args.outfile, args.nshaves, args.inputsize, args.weights)
File "/usr/local/bin/mvNCCompile", line 104, in create_graph
net = parse_tensor(args, myriad_config)
File "/usr/local/bin/ncsdk/Controllers/TensorFlowParser.py", line 700, in parse_tensor
args=arguments)
File "/usr/local/bin/ncsdk/Models/NetworkStage.py", line 279, in init
self.outputDimX))).astype(
ValueError: negative dimensions are not allowe

Comments

Sign In or Register to comment.