Unpooling layer in tensorRT

I am trying to implement semantic semnentation network using tensorRT 3.0.1RC API. But i couldn’t find Unpooling layer in the API. Can you please tell me how can i write code for Unpooling using the API ?

Hi,

Which framework do you use?
unpooling layer is supported in TensorRT 3.0 but FLOAT mode only.

Thanks.

Thanks for replying.
I am using TensorRT 3.0 C++ API.
Can you kindly share the link to Unpooling layer in API ?
Thanks

lovekesh,
So which framework you are using? Caffe, TensorFlow or …

Hi Chijen
Thanks for replying.
I am using neither of them. I am using the tensorRT API without using any parser for caffe or tensorflow.

Hi, lovekesh

After checking, this layer is only for internal usage. Sorry about this.
There is no available document/information to share yet.

An alternative is to implement it with plugin API.
Since you are building network with native API, you can add a plugin layer for unpooling directly:
http://docs.nvidia.com/deeplearning/sdk/tensorrt-api/topics/classnvinfer1_1_1_i_network_definition.html#a4b31ce474ba291e1c95e39ab13c76942

Please let us know if you need further help.
Thanks.

Thanks AastaLLL for replying.
I am already trying to write my own Plugin for max pooling to better understand the entire process of writing plugin. However i am stuck with a segmentation fault for 4 days. I have gone through all the code samples available but of no use.

Can you kindly take a look at this thread (linked below), where i have posted my code snippets and cuda-gdb dump.

https://devtalk.nvidia.com/default/topic/1029951/jetson-tx1/add-custom-maxpool-layer-using-ipluginfactory-/post/5239809/

Many Thanks

Sure.

Let’s track your issue on topic 1029951:
https://devtalk.nvidia.com/default/topic/1029951/jetson-tx1/add-custom-maxpool-layer-using-ipluginfactory-/

Thanks.