Onnx2tf

- `Gather`
- Support for param replacement
- [Weird bug in RoiAlign 22](https://github.com/PINTO0309/onnx2tf/issues/22)

|No.|OP type|Remarks|
|:-:|:-|:-|
|6|Gather|1. "param_target": "inputs"<br>`values`: Value of `indices`<br>`pre_process_transpose_perm`: Transpose is applied to the tensor before the Reshape operation with the perm specified as pre-processing.<br>2. "param_target": "outputs"<br>`post_process_transpose_perm`: Transpose is applied to the tensor after the Reshape operation with the perm specified as post-processing.|

What's Changed
* model convert result wiki test by PINTO0309 in https://github.com/PINTO0309/onnx2tf/pull/24
* Change in method of obtaining onnx filename and change to actions/setup-pythonv4 by PINTO0309 in https://github.com/PINTO0309/onnx2tf/pull/25


**Full Changelog**: https://github.com/PINTO0309/onnx2tf/compare/1.1.34...1.1.35

- Modified the `tests` folder so that it is not included in the installer
- Reviewed CI execution conditions
- Automatically run conversion tests for 42 different models when pull requests are received
|No.|Model|Pass|
|:-:|:-|:-:|
|1|age_googlenet.onnx|:heavy_check_mark:|
|2|arcfaceresnet100-8.onnx|:heavy_check_mark:|
|3|baseline_simplified.onnx|:heavy_check_mark:|
|4|bvlcalexnet-12.onnx|:heavy_check_mark:|
|5|caffenet-12.onnx|:heavy_check_mark:|
|6|densenet-12.onnx|:heavy_check_mark:|
|7|digits.onnx|:heavy_check_mark:|
|8|efficientformer_l1.onnx|:heavy_check_mark:|
|9|efficientnet-lite4-11_nchw.onnx|:heavy_check_mark:|
|10|effnet_opset11_dynamic_axis.onnx|:heavy_check_mark:|
|11|emotion-ferplus-8_rename.onnx|:heavy_check_mark:|
|12|face_detection_yunet_2022mar.onnx|:heavy_check_mark:|
|13|face_recognition_sface_2021dec-act_int8-wt_int8-quantized.onnx|:heavy_check_mark:|
|14|face_recognition_sface_2021dec.onnx|:heavy_check_mark:|
|15|gender_googlenet.onnx|:heavy_check_mark:|
|16|inception-v2-9.onnx|:heavy_check_mark:|
|17|mobilenetv2-12.onnx|:heavy_check_mark:|
|18|mosaic_11.onnx|:heavy_check_mark:|
|19|mosaic-9.onnx|:heavy_check_mark:|
|20|movenet_multipose_lightning_192x256_p6.onnx|:heavy_check_mark:|
|21|nanodet-plus-m_416.onnx|:heavy_check_mark:|
|22|object_tracking_dasiamrpn_kernel_cls1_2021nov.onnx|:heavy_check_mark:|
|23|object_tracking_dasiamrpn_kernel_r1_2021nov.onnx|:heavy_check_mark:|
|24|object_tracking_dasiamrpn_model_2021nov.onnx|:heavy_check_mark:|
|25|qlinear_conv_tensor_test.onnx|:heavy_check_mark:|
|26|rcnn-ilsvrc13-9.onnx|:heavy_check_mark:|
|27|regnet_x_400mf.onnx|:heavy_check_mark:|
|28|ResNet101-DUC-12.onnx|:heavy_check_mark:|
|29|resnet18-v1-7.onnx|:heavy_check_mark:|
|30|resnet50-v1-12.onnx|:heavy_check_mark:|
|31|resnet50-v2-7.onnx|:heavy_check_mark:|
|32|retinanet-9.onnx|:heavy_check_mark:|
|33|squeezenet1.0-12.onnx|:heavy_check_mark:|
|34|super-resolution-10.onnx|:heavy_check_mark:|
|35|tinyyolov2-8.onnx|:heavy_check_mark:|
|36|version-RFB-640.onnx|:heavy_check_mark:|
|37|yolact_edge_mobilenetv2_550x550.onnx|:heavy_check_mark:|
|38|yolov7_tiny_head_0.768_post_480x640.onnx|:heavy_check_mark:|
|39|yolox_nano_192x192.onnx|:heavy_check_mark:|
|40|yolox_nano_416x416.onnx|:heavy_check_mark:|
|41|yolox_s.onnx|:heavy_check_mark:|
|42|zfnet512-12.onnx|:heavy_check_mark:|

What's Changed
* Implement tests for model transformations by PINTO0309 in https://github.com/PINTO0309/onnx2tf/pull/23


**Full Changelog**: https://github.com/PINTO0309/onnx2tf/compare/1.1.33...1.1.34

- `NonMaxSuppression`
- Bug fixes. `max_output_boxes_per_class`
- Support for `Unique`

What's Changed
* Unique operation implementation, bug fix in NonMaxSuppression by Hyunseok-Kim0 in https://github.com/PINTO0309/onnx2tf/pull/21


**Full Changelog**: https://github.com/PINTO0309/onnx2tf/compare/1.1.32...1.1.33

- Improved `Erf` calculation error
- https://github.com/PINTO0309/onnx2tf/releases/download/1.1.28/efficientformer_l1.onnx
python
eps = 1e-11
x_abs = tf.math.abs(input_tensor)
sign = tf.math.divide(input_tensor, tf.math.abs(input_tensor)+eps)
a1 = 0.254829592
a2 = -0.284496736
a3 = 1.421413741
a4 = -1.453152027
a5 = 1.061405429
p = 0.3275911
t = 1.0/(1.0 + p*x_abs)
y = 1.0 - (((((a5*t + a4)*t) + a3)*t + a2)*t + a1)*t*tf.math.exp(-x_abs*x_abs)
erf_tensor = sign*y
tf_layers_dict[graph_node_output.name]['tf_node'] = erf_tensor


**Full Changelog**: https://github.com/PINTO0309/onnx2tf/compare/1.1.31...1.1.32

- Support for `--replace_erf_to_pseudo_erf` option
- Replace `Erf` not supported by TFLite with a primitive approximation
- https://github.com/PINTO0309/onnx2tf/releases/download/1.1.28/efficientformer_l1.onnx
- approximate expression
python
eps = 1e-11
x_abs = tf.math.abs(input_tensor)+eps
sign = tf.math.divide(input_tensor, x_abs)
a1 = 0.254829592
a2 = -0.284496736
a3 = 1.421413741
a4 = -1.453152027
a5 = 1.061405429
p = 0.3275911
t = 1.0/(1.0 + p*x_abs)
y = 1.0 - (((((a5*t + a4)*t) + a3)*t + a2)*t + a1)*t*tf.math.exp(-x_abs*x_abs)
erf_tensor = sign*y
tf_layers_dict[graph_node_output.name]['tf_node'] = erf_tensor


**Full Changelog**: https://github.com/PINTO0309/onnx2tf/compare/1.1.30...1.1.31

- Modified criteria for ShuffleNet patterns to be more strict
- https://github.com/PINTO0309/onnx2tf/releases/download/1.1.30/nanodet-plus-m_416.onnx

**Full Changelog**: https://github.com/PINTO0309/onnx2tf/compare/1.1.29...1.1.30