statistics.py

import torch
from evidential.models import *
from evidential.save import *
from models import *

from torchviz import make_dot




def std_prob(probabilities):

    # Berechnen Sie die Standardabweichung entlang der Dimension 1
    std_dev = torch.std(probabilities, dim=1)

    return std_dev

def divide_by_total(evidential_outputs):
    total = evidential_outputs["total"]
    aleatoric_1_by_total = evidential_outputs["aleatoric_1"]/total
    epistemic_1_by_total = evidential_outputs["epistemic_1"]/total
    aleatoric_2_by_total = evidential_outputs["aleatoric_2"]/total
    epistemic_2_by_total = evidential_outputs["epistemic_2"]/total
    return aleatoric_1_by_total, epistemic_1_by_total, aleatoric_2_by_total, epistemic_2_by_total

def visualize_torchviz():
    dummy_evidential_model = EvidentialWrapper().cuda()
    dummy_aarmvsnet_model = AARMVSNetWrapper().cuda()

    dummy_input = torch.randn(1, 32, 128, 160).cuda()
    output = dummy_evidential_model(dummy_input)
    dot = make_dot(output, params=dict(dummy_evidential_model.named_parameters()))
    dot.render("/home/grannemann/Downloads/evidential_model", format="png")

    dummy_input = torch.randn(1, 5, 3, 128, 160).cuda()
    output = dummy_aarmvsnet_model(dummy_input)
    dot = make_dot(output, params=dict(dummy_evidential_model.named_parameters()))
    dot.render("/home/grannemann/Downloads/aarmvsnet_model", format="png")

#TODO AARMVSNet broken
def export_onnx():
    dummy_evidential_model = EvidentialWrapper().cuda()
    dummy_aarmvsnet_model = AARMVSNetWrapper().cuda()

    dummy_input = torch.randn(1, 32, 128, 160).cuda()
    torch.onnx.export(dummy_evidential_model, dummy_input, '/home/grannemann/Downloads/evidential_model.onnx', input_names=["Features"], output_names=["Evidential Parameters"], opset_version=11)
    print("Exported Evidential onnx model.")

    dummy_input = torch.randn(1, 5, 3, 128, 160).cuda()
    torch.onnx.export(dummy_aarmvsnet_model, dummy_input, '/home/grannemann/Downloads/aarmvsnet_model.onnx', input_names=["image"], output_names=["logits"], opset_version=14, verbose=True)
    print("Exported AARMVSNet onnx model.")


if __name__ == "__main__":
    export_onnx()