Intel® Distribution of OpenVINO™ Toolkit
Community assistance about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all aspects of computer vision-related on Intel® platforms.

how can use 5d placeholder for NCS2 on openvino

lin__weicheng
Beginner
332 Views

I convert the 3dcnn model success, but if I use this model on openvino, it says ['RuntimeError: [VPU] Convolution supports only 3D or 4D input].

The model placeholder shape is [batch, depth, height, width, channel]

If NCS2 can use this placeholder in model? and how can I do?

0 Kudos
3 Replies
Shubha_R_Intel
Employee
332 Views

Dear lin, weicheng,

Can you convert your 5D model to a 3 or 4D model before you run it through Model Optimizer ? It should be possible to re-design your model so that it doesn't have to use 5D . A 5D model inferenced on MYRIAD  would almost not be worth it - it would be very slow. This is why only 3D or 4D conv layers is supported on MYRIAD.

Thanks,

Shubha

0 Kudos
lin__weicheng
Beginner
332 Views

Dear Shubha,

I think it can't, because I use the tensorflow conv3d function, this function inputs with shape (batch, depth, height, width, channels).

I try to convert this, it errors [ Input 0 of layer conv3d_1 is incompatible with the layer: expected ndim=5, found ndim=4].
Maybe is I don't know how can I do, so can you provide a example to convert dimention.

Thanks

0 Kudos
Shubha_R_Intel
Employee
332 Views

Dear lin, weicheng,

The error is pretty clear.  

 ['RuntimeError: [VPU] Convolution supports only 3D or 4D input].

Please see this tf.reshape issue . Maybe you need to flatten the Placeholder into a vector then resize it to 3D or 4D. 

Hope it helps,

Thanks,

Shubha

0 Kudos
Reply