I’m very glad to see there is a offical mask rcnn tensorRT sample with well-known matterport’s keras implement on https://github.com/NVIDIA/TensorRT/tree/master/samples/opensource/sampleUffMaskRCNN.
But Could you tell us that if it only works with TensorRT6.0? but can’t work with TensorRT5.15 on Jetson?
Cause i have encounter the same problem as in [REFERENCE] KeyError: 'mrcnn_mask_bn4/batchnorm/mul_1' in running sampleUffMaskRCNN demo · Issue #123 · NVIDIA/TensorRT · GitHub
If it can’t run on jetson with tensorRT5.15,should I cross-compile it on PC with tensorrt6.0,and put it run on jetson?
By the way, this code maybe is quit problematic,could nvidia engineers refine it to make it more “offical” soon?
Thanks alot!
ResizeBilinear is added in TRT 6.0 for tensorflow which is required for mask rcnn.
Serialized engines are not portable across platforms or TensorRT versions. Engines are specific to the exact GPU model they were built on (in addition to the platforms and the TensorRT version).
Regarding TRT 6.0 support on Jetson:
JetPack 4.3 Developer Preview, which is packaged with TRT6 is available as a beta. JetPack 4.3 Developer Preview | NVIDIA Developer
Stay tuned for official production release of Jetpack 4.3.
Hi there, I am currently doing conversion and running TRT 7.1 so that I can run this tensorRT version of the model on a Jetson Xavier AGX. I am planning to follow the instruction as stated in https://github.com/NVIDIA/TensorRT/tree/master/samples/opensource/sampleUffMaskRCNN for the conversion on my Host PC (Ubuntu 18.04 with TRT 7.1 installed)
I planning to run inference on Jetson Xavier AGX installed with jetpack 4.4.1 which has TRT 7.1 installed on it. If I used TRT 6.0 for the conversion on my Host PC, can it work on TRT 7.1? I check and TRT 7.1 has the support for ResizeBilinear.