TensorRT 1.0 does not support Caffe MVNLayer

dividiti · January 9, 2017, 9:58am

It appears that the TensorRT 1.0 parser does not support the normalisation layer type: http://caffe.berkeleyvision.org/doxygen/classcaffe_1_1MVNLayer.html reporting:

could not parse layer type MVN

Any plans to support this in TensorRT 2.0?

dividiti · January 18, 2017, 1:19pm

According to a reply from NVIDIA, this has been added as a feature request (although not guaranteed to be implemented).

Topic		Replies	Views
TensorRT- unable to parse layers in SSD_VGG network TensorRT	3	1308	March 15, 2019
Parsing a caffe normalize layer TensorRT	2	1216	May 23, 2019
Does TensorRT(GIE) support Batch Normalization? GPU-Accelerated Libraries	2	1526	January 12, 2018
Convert TensorRT output to Caffe GPU-Accelerated Libraries	0	871	November 21, 2016
Slice layer is not supported by nvcaffeparser1::ICaffeParser (TensorRT 5.1) TensorRT	2	1543	May 16, 2019
Could not parse layer type Normalize TensorRT	3	665	November 17, 2021
Caffe parser has no feedback at all. TensorRT	2	441	October 12, 2021
Using DetectNet caffe model in sample_object_detector DriveWorks	3	2525	February 13, 2018
tensorRT doesn't support PReLu ? Jetson TX2	2	1475	October 18, 2021
TensortRT vs NVCaffe General	1	3589	October 16, 2017