tensorflow-onnx
tensorflow-onnx copied to clipboard
onnx
Convert TensorFlow, Keras, Tensorflow.js and Tflite models to ONNX
# Ask a Question Hello all，when I convert a tensorflow QAT model into onnx with tf2onnx, there is a error: `raise ValueError("make_sure failure: " + error_msg % args) ValueError: make_sure...
**Describe the bug** Converting `tf.keras.backend.resize_images` with interpolation mode `nearest` to ONNX and then to TensorRT produces mismatching results. The created `Resize` op uses rounding method `floor` as default. Although `round_prefer_floor`...
# Ask a Question ### I have just started learning how to convert a TensorFlow model to ONNX. How should I start fix this error? ERROR - Tensorflow op [cudnn_lstm/rnn/multi_rnn_cell/cell_0/cudnn_compatible_lstm_cell/BlockLSTM:...
**Describe the bug** I have 2 signature to export to ONNX from same model. After conversion, 1 of them is fine, same results of TF and ONNX. But another signature...
I use tf2onnx.convert.from_keras to transform my yolo keras model to onnx but the sequence of onnx output is not the same as model output (ex: model: 8*13*80, 16*26*80, 32*52*80; onnx:8*13*80,...
Convert quantization aware trained model from TF to ONNX has several issues -- 1. `QuantizeLinear` and `DequantizeLinear` are fused into conv layer, but the downstream compiler(e.g., TensorRT) needs the Q/DQ...
# Converting Keras Saved Model to ONNX using a restricted set of operators ### Question Hi Team! First of all, thanks for this wonderful tool! My question is - Is...
**Describe the bug** Conversion of TensorFlow model ([TF-Hub](https://tfhub.dev/google/universal-sentence-encoder-multilingual-large/3)) to ONNX failed **Urgency** a project deployment deadline 7th April 2023 **System information** - OS Platform and Distribution (e.g., Linux Ubuntu 18.04*):...
**Describe the bug** The library fails to export a model that uses a LSTM layer within a loop. This was relevant for encoding an autoregressive generation loop within an ONNX...
**Describe the bug** Currently, we need a `ceildiv` operation to compute the sequence lengths after convolution for sequence models. The `ceildiv` operation is implemented as ``` ceildiv(a, b) := -floordiv(-a,...
