How to build tensorflow from source with a model that uses custom ops which are renamed versions of existing ops? - tensorflow

Question

I have a .tflite model that uses custom operations called MaxPoolingWithArgmax2D, MaxUnpooling2D, and Convolution2DTransposeBias.
These ops are actually not custom ops, since they are present in tensorflow already (MaxPoolWithArgmax, MaxUnpooling2D, conv2d_transpose)
After consulting this guide, I see that I'd have to write a kernel and an interface for these ops.
Is there a way to build tensorflow source without writing custom implementations for these ops since they're already present in the library? The only problem is that the model I'm using has renamed them due to which they're being recognized as custom ops. My goal is to perform inference with this model.
Edit: These ops are not select ops. They are built-in ops present inside the base library. However, the person who wrote this model renamed them, which makes them custom ops.
Edit 2: Photo for reference:

Answer 1

You can enable the existin TF ops through the Select TF option in the TFLite.
For example, during the conversion stage, you can enable them:
converter.target_spec.supported_ops = [
tf.lite.OpsSet.TFLITE_BUILTINS, # enable TensorFlow Lite ops.
tf.lite.OpsSet.SELECT_TF_OPS # enable TensorFlow ops.
]
For the inference stage, please make sure that the select tf dependency is linked. When using TF Python API, it will automatically be enabled.
Please refer to this link.
Among the custom ops, the custom operator implementations for some of them are being distributed as a perception operator package.
from tensorflow.lite.kernels.perception import pywrap_perception_ops as perception_ops_registerer
from tensorflow.lite.python import interpreter as interpreter_wrapper
interpreter = interpreter_wrapper.InterpreterWithCustomOps(
model_content=model,
custom_op_registerers=[
perception_ops_registerer.PerceptionOpsRegisterer
])
Please take a look at this link.