Is it able to only use onnxruntime for inference after conversion through mmdeploy?

[kaizengit]

For example, I've used mmdeploy to convert a binary classification model from torch to onnx.

python mmdeploy/tools/deploy.py \
    mmdeploy/configs/mmpretrain/classification_onnxruntime_static.py
     ......

It succussed and when I use the Backend model inference code in doc it outputs correct result.

# read deploy_cfg and model_cfg
deploy_cfg, model_cfg = load_config(deploy_cfg, model_cfg)

# build task and backend model
task_processor = build_task_processor(model_cfg, deploy_cfg, device)
model = task_processor.build_backend_model(backend_model)

# process input image
input_shape = get_input_shape(deploy_cfg)
model_inputs, _ = task_processor.create_input(image, input_shape)

# do model inference
with torch.no_grad():
    result = model.test_step(model_inputs)

I thought onnx models could be independently run with onnxruntime, like this:

session = onnxruntime.InferenceSession(onnx_path, providers=['CUDAExecutionProvider'])
img_array = Image.open(img_name).resize((256, 256))

img_array = np.array(img_array).astype('float32')
img_array = np.transpose(img_array, (2, 0, 1))
img_array = np.expand_dims(img_array, axis=0)  

）
input_name = session.get_inputs()[0].name
output_name = session.get_outputs()[0].name

result = session.run([output_name], {input_name: img_array})[0]

print("Raw output:", result)

However, the output always equal to [[0., 1.]]. Any idea to simply use onnxruntime for inference? Is that because some processing or configuration in the task_processor?