agrobot_base/Python/OAK/datasets/_7_convert_deeplabv3.py

import os
import torch
from torchvision.models.segmentation import deeplabv3_resnet50

MODELO = "oak-1"
MODEL_NAME = "ervasModel"
NUM_CLASSES = 3
RESOLUCAO = (512, 512)
model_folder = os.path.join(MODELO, "backup")
model_name = MODEL_NAME + "_best"

# Carrega seu modelo (usa num_classes igual ao treino)
model = deeplabv3_resnet50(weights=None, num_classes=NUM_CLASSES)
model.load_state_dict(torch.load(os.path.join(model_folder, f"{model_name}.pth")))
model.eval()

# Dummy input com batch_size=1, channels=3, height=512, width=512
dummy_input = torch.randn(1, 3, RESOLUCAO[0], RESOLUCAO[1])

# Exporta
torch.onnx.export(
    model,
    dummy_input,
    os.path.join(model_folder, f"{model_name}.onnx"),
    input_names=["input"],
    output_names=["output"],
    opset_version=11
)