search

davidsandberg / facenet

Face recognition using Tensorflow
MIT License
13.84k stars 4.81k forks source link

Unable to convert onnx model to TRT model #1245

Open k-khosravi opened 11 months ago

k-khosravi commented 11 months ago

I have converted this model "20180402-114759" for face recognition to onnx format with onnx==1.14.1. and When I want to convert the onnx model to TRT I get these errors: "[E] Error[4]: [graphShapeAnalyzer.cpp::processCheck::862] Error Code 4: Internal Error (StatefulPartitionedCall/inception_resnet_v1/Conv2d_2a_3x3/Conv2D: spatial dimension of convolution/deconvolution output cannot be negative (build-time output dimension of axis 2 is -2)) [12/16/2023-15:52:45] [E] Engine could not be created from network [12/16/2023-15:52:45] [E] Building engine failed [12/16/2023-15:52:45] [E] Failed to create engine from model or file. [12/16/2023-15:52:45] [E] Engine set up failed &&&& FAILED TensorRT.trtexec [TensorRT v8601] # ...." I use TensorRT 8.6 for CUDA 12.1. This doc said that for tensorrt 8.6 the onnxruntime should be 1.15 or higher. but I am not able to convert the tensorflow model to onnx model with onnx==1.15.0. and I face some errors. How can I fix this problem to successfully convert the tensorflow model to onnx and then TRT model. Best Regards!

K-prog commented 3 months ago

@k-khosravi can you provide the code you used for onnx conversion?