1.21.1

• Constant

  • Bring Constant layers unconnected to the model into the model.

  • It is assumed that the -nuo option is specified because running onnxsim will remove constants from the ONNX file.

  • Wrap constants in a Lambda layer and force them into the model.

  • toy_with_constant.onnx.zip

  • Convert test

    onnx2tf -i toy_with_constant.onnx -nuo -cotof

    ONNX	TFLite

    

  • Inference test

    import tensorflow as tf
    import numpy as np
    from pprint import pprint
    
    interpreter = tf.lite.Interpreter(model_path="saved_model/toy_with_constant_float32.tflite")
    interpreter.allocate_tensors()
    
    input_details = interpreter.get_input_details()
    output_details = interpreter.get_output_details()
    
    interpreter.set_tensor(
        tensor_index=input_details[0]['index'],
        value=np.ones(tuple(input_details[0]['shape']), dtype=np.float32)
    )
    interpreter.invoke()
    
    variable_output = interpreter.get_tensor(output_details[0]['index'])
    constant_output = interpreter.get_tensor(output_details[1]['index'])
    
    print("=================")
    print("Variable Output:")
    pprint(variable_output)
    print("=================")
    print("Constant Output:")
    pprint(constant_output)

    =================
    Variable Output:
    array([[-0.02787317, -0.05505124,  0.05421712,  0.03526559, -0.14131774,
             0.0019211 ,  0.08399964,  0.00433664, -0.00984338, -0.03370604]],
          dtype=float32)
    =================
    Constant Output:
    array([1., 2., 3., 4., 5.], dtype=float32)
    

• Constant outputs removed from ONNX during conversion #627

1.21.0

• Fix Typo
• Change API parameter names

What's Changed

• Fix Typo by @PINTO0309 in #625

Full Changelog: 1.20.10...1.21.0

1.20.10

• YOLOvN ONNX
• Fixed to restore metadata
• pip install -U sng4onnx>=1.0.4 sne4onnx>=1.0.13
  

What's Changed

• Fixed to restore metadata by @PINTO0309 in #624

Full Changelog: 1.20.9...1.20.10

1.20.9

• Tile, Reshape
  • Tensor processing of dynamic shapes is supported.
  • Due to the experimental implementation, regression testing is still insufficient. However, this ONNX is ready for conversion.
  • daquexian/onnx-simplifier#323
  • src_ocr.onnx.zip
  • https://github.com/PaddlePaddle/PaddleOCR
  • Input: [N, 3, 32, W]]
    
    
    

What's Changed

• [experimental] Support for dynamic Tile, dynamic Reshape by @PINTO0309 in #623

Full Changelog: 1.20.8...1.20.9

1.20.8

• If
  • Improved conversion stability of subgraphs of If operations.
  • onnx/onnx-tensorflow#1056
  • https://github.com/PINTO0309/onnx2tf/releases/download/1.20.7/maskrcnn_resnet50_fpn.onnx

onnx2tf \
-i maskrcnn_resnet50_fpn.onnx \
-onimc boxes.55 onnx::Shape_3316 3315 onnx::Loop_3751

What's Changed

• Improved conversion stability of subgraphs of If operations. by @PINTO0309 in #622

Full Changelog: 1.20.7...1.20.8

1.20.7

• Optimization of torch.nn.PixelUnshuffle.

  • https://pytorch.org/docs/stable/generated/torch.nn.PixelUnshuffle.html
  • https://www.tensorflow.org/api_docs/python/tf/nn/space_to_depth
  • This implementation is inspired by the valuable knowledge of AlexanderLutsenko / nobuco. Thank you very much.
    • https://github.com/AlexanderLutsenko/nobuco?tab=readme-ov-file#implementation-mismatch-pick-your-poison
  • Unlike PyTorch's method of replacing internal processing, a similar process is implemented from a frozen ONNX graph with the minimum number of operations required.

  ONNX	Before tflite	After tflite

• Bug fix. ReduceL1, ReduceL2, ReduceLogSum, ReduceLogSumExp, ReduceMax, ReduceMean, ReduceMin, ReduceProd, ReduceSum, ReduceSumSquare

What's Changed

• Optimization of torch.nn.PixelUnshuffle by @PINTO0309 in #621

Full Changelog: 1.20.6...1.20.7

1.20.6

• MaxPool, AveragePool
  • Improved conversion stability when H, W and D of MaxPool and AveragePool contain undefined dimensions.
  • The accuracy of the converted model is not always accurate.
  • e.g. YOLOvNn dynamic inputs [N, 3, H, W]
  • YOLOvN has been modified only to avoid a situation where the conversion aborts, although this is undoubtedly not good for the design of the model, as fixed parameters such as the number of classes are embedded in the backward Split operation with fixed values.
  • Concatenating dimensions that have completely different meanings is also a major problem.
  • Unless you replace the PyTorch implementation with Slice, you won't be able to do proper inferencing.
    • ONNX
      
      
    • TFLite
      
      

What's Changed

• Improved conversion stability when H, W and D of MaxPool and AveragePool contain undefined dimensions by @PINTO0309 in #620

Full Changelog: 1.20.5...1.20.6

1.20.5

• Distribute test data files to GitHub releases.
  • calibration_image_sample_data_20x128x128x3_float32.npy
  • https://github.com/PINTO0309/onnx2tf/releases/download/1.20.4/calibration_image_sample_data_20x128x128x3_float32.npy
  • connection timeout: 1 sec
  • read timeout: 5 sec
• The downstream speed of Wasabi Storage is very slow due to the rapid increase in the number of requests per day.
• Porting to Wasabi Storage if the download from GitHub releases times out.

1.20.4

• [experimental, Breaking change] tf.keras -> tf_keras
• pip install tf-keras~=2.16
• Because the API path for tf.keras is obsolete.

What's Changed

• [experimental, Breaking change] tf.keras -> tf_keras by @PINTO0309 in #618

Full Changelog: 1.20.3...1.20.4

1.20.3

• ReduceL1, ReduceL2, ReduceLogSum, ReduceLogSumExp, ReduceMax, ReduceMean, ReduceMin, ReduceProd, ReduceSum, ReduceSumSquare
  • Improve stability of axes conversion when an unknown dimension order other than NCHW is input to ReduceXXX.
  • Automatically compensates for mistakes in axes dimensional transpositions.
• Error
  
  
• Fixed
  
• Op Error about "Transpose" #616

What's Changed

• Improve stability of axes conversion when an unknown dimension order other than NCHW is input to ReduceXXX by @PINTO0309 in #617

Full Changelog: 1.20.2...1.20.3