Onnx2tf

- Significantly improved processing efficiency when interrupting model transformation at an arbitrary position.

What's Changed
* Significantly improved processing efficiency when interrupting model transformation at an arbitrary position by PINTO0309 in https://github.com/PINTO0309/onnx2tf/pull/112


**Full Changelog**: https://github.com/PINTO0309/onnx2tf/compare/1.5.3...1.5.4

- `--check_onnx_tf_outputs_elementwise_close_full` option adds ability to verify errors in order from model entry

https://user-images.githubusercontent.com/33194443/211808556-554dd54c-1cef-48e3-a55d-357db50ce26d.mp4

What's Changed
* `--check_onnx_tf_outputs_elementwise_close_full` option adds ability to verify errors in order from model entry by PINTO0309 in https://github.com/PINTO0309/onnx2tf/pull/111


**Full Changelog**: https://github.com/PINTO0309/onnx2tf/compare/1.5.2...1.5.3

- `--check_onnx_tf_outputs_elementwise_close`
- Change initial values for relative and absolute tolerances
- `rtol`
- `1e-5` -> `0.0`
- `atol`
- `1e-5` -> `1e-4`

What's Changed
* Change initial values for relative and absolute tolerances by PINTO0309 in https://github.com/PINTO0309/onnx2tf/pull/110


**Full Changelog**: https://github.com/PINTO0309/onnx2tf/compare/1.5.1...1.5.2

- `Sub`, `Div`, `Mod`
- Fixed a bug in `pre_explicit_broadcast` that caused the order of operations to be reversed for `Sub`, `Div`, and `Mod`.
- `MaxPool`
- Fixed padding bug

What's Changed
* Fixed a bug in pre_explicit_broadcast that caused the order of operations to be reversed for Sub, Div, and Mod. by PINTO0309 in https://github.com/PINTO0309/onnx2tf/pull/109


**Full Changelog**: https://github.com/PINTO0309/onnx2tf/compare/1.5.0...1.5.1

- [Experimental] Added the ability to validate the model final output tensor in ONNX and TensorFlow.
https://numpy.org/doc/stable/reference/generated/numpy.allclose.html#numpy-allclose

numpy.allclose(a, b, rtol=1e-05, atol=1e-05, equal_nan=True)

- This option in combination with the `--output_names_to_interrupt_model_conversion` option can be used to investigate which operations at which locations in the model cause errors in the output.
- Since ONNX assumes NCHW and TensorFlow assumes NHWC output, a simple comparison of output tensors will not match values in most cases. Therefore, the tool automatically tries to match the final output tensor of TensorFlow to the shape of the output tensor of ONNX with a brute force check. If the shape still does not match or there is no exact matching value combination, `Unmatched` is assumed.
- Currently, the function only provides validation with the output OP specified by the user, but eventually a function will be added to automatically search for OPs with large output errors.

-coto, --check_onnx_tf_outputs_elementwise_close
Returns true if the two arrays, the output of onnx and the output of TF,
are elementwise close within an acceptable range.

-cotor CHECK_ONNX_TF_OUTPUTS_ELEMENTWISE_CLOSE_RTOL,\
--check_onnx_tf_outputs_elementwise_close_rtol CHECK_ONNX_TF_OUTPUTS_ELEMENTWISE_CLOSE_RTOL
The relative tolerance parameter.
Default: 1e-5

-cotoa CHECK_ONNX_TF_OUTPUTS_ELEMENTWISE_CLOSE_ATOL,\
--check_onnx_tf_outputs_elementwise_close_atol CHECK_ONNX_TF_OUTPUTS_ELEMENTWISE_CLOSE_ATOL
The absolute tolerance parameter.
Default: 1e-5


- e.g.

onnx2tf -i xxx.onnx -coto -onimc keypoints descriptors scores scores_map



What's Changed
* [Experimental] Added the ability to validate the model final output tensor in ONNX and TensorFlow by PINTO0309 in https://github.com/PINTO0309/onnx2tf/pull/106


**Full Changelog**: https://github.com/PINTO0309/onnx2tf/compare/1.4.2...1.5.0

- `Squeeze`, `Unsqueeze`
- Only when `squeeze` and `unsqueeze` are consecutive and each is performing a useless process of compressing and decompressing the same `axis`, the two operations are disabled at the same time. This is an optimization to avoid as much as possible the risk of shape irregularities in the next connected operation of `Unsqueeze`.
- https://github.com/PINTO0309/onnx2tf/releases/download/1.1.28/ppmattingv2_stdc1_human_480x640.onnx

What's Changed
* Only when `Squeeze` and `Unsqueeze` are consecutive and each is performing a useless process of compressing and decompressing the same `axis`, the two operations are disabled at the same time. by PINTO0309 in https://github.com/PINTO0309/onnx2tf/pull/102


**Full Changelog**: https://github.com/PINTO0309/onnx2tf/compare/1.4.1...1.4.2