Cast operation not working

Applying a node Cast with targetType uint8 on a float tensor is not working.
Here is the test script:

import aidge_core; import aidge_backend_cpu; import aidge_onnx
import numpy as np

cast_node = aidge_core.Cast(aidge_core.dtype.uint8, "cast_node")
cast_graph = aidge_core.get_connected_graph_view(cast_node)

X = np.random.rand(1, 3, 3).astype(np.float32) * 255  # Scale to match UINT8 range
x_tensor = aidge_core.Tensor(X)

cast_graph.compile("cpu", aidge_core.dtype.float32, dims=[[1, 3, 3]])
scheduler = aidge_core.SequentialScheduler(cast_graph)

scheduler.forward(data=[x_tensor])
for outNode in cast_graph.get_output_nodes():
	output_aidge = np.array(outNode.get_operator().get_output(0))
	print("Aidge output: \n",output_aidge)

The output is still float.

added StatusWork in Progress label

assigned to @hrouis

added Bug 🐛 TopicOperator labels

changed milestone to %aidge v0.7.0

changed the description

Update:
At node creation the output is well set to the target type, I checked.
However, when doing the forward, the output is not of target type (it is back to float in my case).
So somehow, the type of the output is overwritten again by the wrong original type.

mentioned in merge request !362 (merged)

The issue comes from setDataType() of OperatorTensor that overwrites the output type with the input's type (bad type) after setting with the target type (good type) by the constructor.
A fix is added in !362 (merged)

added StatusReview Ready label and removed StatusWork in Progress label

Actually, your test code is questionnable: you should not use the compile method, which sets the data type of the graph (we should really rename this method by the way!). Then, the question is, can the user change the data type of Cast operator or should it remains constant from the construction?

closed