1.26.7

• DequantizeLinear
  • Fixed the broadcast processing when x_scale is 1D 1Elem.
  • best.onnx.zip
  • onnx2tf -i best.onnx -ois images:1,3,512,640
    
    
• DequantizeLinear.py can't compile layer when x_scale_rank == 1 #733

What's Changed

• Fixed the broadcast processing when x_scale is 1D 1Elem by @PINTO0309 in #735

Full Changelog: 1.26.6...1.26.7

1.26.6

• arm64 hosted runner image deploy test
  • #653
  • #659
  • #660
  • #661
• flatc arm64 binary

What's Changed

• arm64 hosted runner image deploy test by @PINTO0309 in #734

Full Changelog: 1.26.5...1.26.6

1.26.5

• NonMaxSuppression

  • Enable selection of V4 and V5 for NonMaxSuppression.

      -snms {v4,v5}, --switch_nms_version {v4,v5}
        Switch the NMS version to V4 or V5 to convert.
        e.g.
        NonMaxSuppressionV4(default): --switch_nms_version v4
        NonMaxSuppressionV5: --switch_nms_version v5
    

    

• TensorFlow.js / tfjs - register_all_kernels.ts

  • wasm

    https://github.com/tensorflow/tfjs/blob/master/tfjs-backend-wasm/src/register_all_kernels.ts#L116-L118

    

  • webgl

    https://github.com/tensorflow/tfjs/blob/master/tfjs-backend-webgl/src/register_all_kernels.ts#L120-L122

    

  • webgpu

    https://github.com/tensorflow/tfjs/blob/master/tfjs-backend-webgpu/src/register_all_kernels.ts#L122-L123

    

What's Changed

• Enable selection of V4 and V5 for NonMaxSuppression by @PINTO0309 in #732

Full Changelog: 1.26.4...1.26.5

1.26.4

• Add
  • #729 fix shape unmatch workaround
    
  • Problems when transforming dynamic input models and quantifying static models #729

What's Changed

• #729 shape unmatch workaround test by @PINTO0309 in #730

Full Changelog: 1.26.3...1.26.4

1.26.3

• MatMul
  Fix incorrect tensor expansion in MatMul operation

1. Content and background

The MatMul operation was incorrectly handling 1-dimensional tensors by expanding
the wrong input tensor. When handling a 1D input tensor (shape [256]), it was
erroneously expanding input_tensor_2 (shape [256, 254]) instead of input_tensor_1,
leading to incorrect shape transformations.

2. Summary of corrections

Changed:

input_tensor_1 = tf.expand_dims(input_tensor_2, axis=0)

to

input_tensor_1 = tf.expand_dims(input_tensor_1, axis=0)

This ensures the correct tensor is expanded when handling 1D inputs.

3. Before/After

Before:

Input1 shape: [256] -> incorrectly became [1,256,254]
Input2 shape: [256,254] remained unchanged

After:

Input1 shape: [256] -> correctly becomes [1,256]
Input2 shape: [256,254] remains unchanged

What's Changed

• renamed replace_GRU.json to allow cloning to Windows by @kwikwag in #718
• Bugfix in MatMul.py by @oesi333 in #725

New Contributors

• @kwikwag made their first contribution in #718
• @oesi333 made their first contribution in #725

Full Changelog: 1.26.2...1.26.3

1.26.2

• Supports multi-batch quantization of image input.

onnx2tf \
-i batch_size_2.onnx \
-oiqt \
-cind images test.npy [[[[0.485,0.456,0.406]]]] [[[[0.229,0.224,0.225]]]]

• Problem of quantization with batch_size 2 #714

What's Changed

• Multi batch quant by @PINTO0309 in #715

Full Changelog: 1.26.1...1.26.2

1.26.1

• Added Float32 as an option for input and output types after quantization.

  -iqd {int8,uint8,float32}, --input_quant_dtype {int8,uint8,float32}
    Input dtypes when doing Full INT8 Quantization.
    "int8"(default) or "uint8" or "float32"
  
  -oqd {int8,uint8,float32}, --output_quant_dtype {int8,uint8,float32}
    Output dtypes when doing Full INT8 Quantization.
    "int8"(default) or "uint8" or "float32"

What's Changed

• Comment fix. input_quant_dtype, output_quant_dtype by @PINTO0309 in #706
• Update replace_slice.json reference by @emmanuel-ferdman in #708
• Fixed mistake of #710 by @marcoschepis in #711
• Added Float32 option by @PINTO0309 in #712

New Contributors

• @emmanuel-ferdman made their first contribution in #708

Full Changelog: 1.26.0...1.26.1

1.26.0

• API changed

  • The input and output quantization types can now be specified with separate, different parameters for input and output.
  • Abolition
    • input_output_quant_dtype
  • Addition
    • input_quant_dtype
    • output_quant_dtype

• Conv

  • Supports parameter substitution for post-processing of Conv.
  • replace_conv.json

    {
      "format_version": 1,
      "operations": [
        {
          "op_name": "wa/conv/Conv",
          "param_target": "outputs",
          "param_name": "output",
          "post_process_transpose_perm": [0,3,1,2]
        }
      ]
    }

  • test

    onnx2tf -i model_conv.onnx -kat input -prf replace_conv.json
    

    

• Mul

  • Improved conversion stability of Mul.
    

• Error by broadcasting in tf.math.Multiply Operation #698

• Add transposition to the tflite model #700

• README corrections due to API changes (I'll get serious from tomorrow) #702

What's Changed

• Input and output quantization chosen separately by @marcoschepis in #701
• Updated README.md by @marcoschepis in #703
• Supports parameter substitution for post-processing of Conv by @PINTO0309 in #704
• Improved conversion stability of Mul by @PINTO0309 in #705

New Contributors

• @marcoschepis made their first contribution in #701

Full Changelog: 1.25.15...1.26.0

1.25.15

• Mul, Add
  • Fixed to force switch between X and Y when X: np.ndarray, Y: Tensor

    onnx2tf -i 1005_s0_nonar_text_decoder.onnx -cotof
    

    
• Error by broadcasting in tf.math.Multiply Operation #698

What's Changed

• Fixed to force switch between X and Y when X: np.ndarray, Y: Tensor by @PINTO0309 in #699

Full Changelog: 1.25.14...1.25.15

1.25.14

• ArgMin
  • Dealing with garbage-like broken structures in ONNX (ArgMin) #695
• [YOLOv7] None in graph_node_input.shape #694

What's Changed

• Dealing with garbage-like broken structures in ONNX (ArgMin) #695 by @PINTO0309 in #696

Full Changelog: 1.25.13...1.25.14