Seeing the following error on a Mask RCNN Tensorflow Saved Model. Using the Triton server container.
F tensorflow/stream_executor/cuda/cuda_dnn.cc:535] Check failed: cudnnSetTensorNdDescriptor(handle_.get(), elem_type, nd, dims.data(), strides.data()) == CUDNN_STATUS_SUCCESS (3 vs. 0)batch_descriptor: {count: 1 feature_map_count: 64 spatial: 0 25 value_min: 0.000000 value_max: 0.000000 layout: BatchDepthYX}
However, if I send an incorrect sized array, I get an error mesage and the trtserver does not crash .
I0410 16:33:36.387566 888 model_repository_manager.cc:511] GetInferenceBackend() 'modelx' version -1
I0410 16:33:36.387604 888 http_server.cc:1214] Infer failed: unexpected shape for input 'image:0' for model 'modelx'. Expected [-1,3], got [75,4]
The output from curl -v localhost:8080/api/status for the model looks as follows.
max_batch_size: 1
input {
name: “image:0”
data_type: TYPE_FP32
dims: -1
dims: 3
}
output {
name: “output/scores:0”
data_type: TYPE_FP32
dims: 1
reshape {
}
}
output {
name: “output/labels:0”
data_type: TYPE_INT64
dims: 1
reshape {
}
}
output {
name: “output/boxes:0”
data_type: TYPE_FP32
dims: 4
}
Any help will be appreciated.