tensorflow-onnx

tensorflow-onnx copied to clipboard

Hello this is a feature request for an --mp "Model Precision" parameter that will influence the output model precision. for example : ``` python3 -m tf2onnx.convert --saved-model tensorflowModel/ --opset 14...

AmmarkoV

Constant in the Reshape operator on s390x architecture (big endian) is a large number 268632064 (expected 784 as on x86)

3

**Dubugging advice** Converting TF model to onnx on s390x succeeds, but the resulting onnx file includes a large number 268632064 in Reshape operator. python3 -m tf2onnx.convert --opset 15 --fold_const --saved-model...

aosadchyy

**Describe the bug** Not sure if it is expected behavior, I was converting a densenet-like model trained by Keras and I notice that it takes tf2onnx over 18 minutes to...

maybeLee

**Describe the bug** Converting a TensorFlow model that accumulates data via a ```tf.TensorArray``` in a nested for-loop *fails*. **Urgency** None **System information** - OS Platform and Distribution: Ubuntu 20.04.4 LTS...

r-remus

Keras - Conv1D conversions with tf2onnx and keras2onnx do not support NWC layout for Conv1D (Keras default)

1

## Support for Conv1D NWC kernel (keras default) I did not find a way to avoid a transposition layer before and after the Conv1D operator and support the NWC kernel...

ed-mr

Cannot export Keras TextVectorization to ONNX

5

Describe the bug A simple model with a TextVectorization layer cannot be converted to ONNX System information OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Ubuntu 20.04 Tf2onnx version: 1.9.3...

hombreola

Failure to assign/calculate groups in Conv3D

1

As seen in the last few posts of #1434. The fix made in #1435 doesn't work for Conv3D. Node group numbers are incorrect because the 5d data formats aren't accounted...

reach-satori

Unsure how to use dequantize flag so that the onnx model can be used with unity barracuda

1

**Describe the bug** receiving message "ValueError: Opset 10 is required for quantization. Consider using the --dequantize flag or --opset 10." After trying to `python -m tf2onnx.convert --opset 9 --dequantize --tflite...

LauraMarieRam

Convert Dense to "FC" instead of "GEMM"

1

Is there a way to convert force the conversion of tf.Densse layer to MatMul + Add instead of GEMM? I'm converting onnx to DLC and GEMM is causing problems.

aviaisr

Gemm Graph Rewriter Fails When Model Has Dynamic Batch Size

1

**Describe the bug** When processing a graph with unspecified batch size, and with a Matmul followed by an Add node, the gemm_rewriter will fail to rewrite the Matmul+Add nodes to...

frank-ff