Error preprocessing Imagenet for GoogleNet: AttributeError: 'NDArray' object has no attribute '__array_interface__'

Hello all, I am trying to evaluate GoogleNet model. I have successfully evaluated other models with the Imagenet dataset. However, GoogleNet requires a different preprocessing, which I cannot reproduce successfully.

# Preprocessing function for ImageNet models using numpy
def transform(img, label):
    '''
    Preprocessing required on the images for inference with mxnet gluon
    The function takes loaded image and returns processed tensor
    '''
    img = np.array(Image.fromarray(img).resize((224, 224))).astype(np.float32)
    img[:, :, 0] -= 123.68
    img[:, :, 1] -= 116.779
    img[:, :, 2] -= 103.939
    img[:,:,[0,1,2]] = img[:,:,[2,1,0]]
    img = img.transpose((2, 0, 1))
    img = np.expand_dims(img, axis=0)

    return img,label

##### Evaluate the entire dataset #############
def evaluate(data_dir, order):
    print("Evaluate")
    ctx = [mx.cpu()]

    # batch size (set to 1 for cpu)
    batch_size = 1

    # number of preprocessing workers
    num_workers = multiprocessing.cpu_count()
    
    val_data = gluon.data.DataLoader(
        imagenet.classification.ImageNet(data_dir, train=False).transform(transform),
        batch_size=batch_size, shuffle=False, num_workers=num_workers)

    ort_session_cpu = ort.InferenceSession(onnx_model.SerializeToString())

    # Compute evaluations
    print("----Running ONNXRuntime----")
    num_batches = int(50000/batch_size)
    print('[0 / %d] batches done'%(num_batches))

    # Loop over batches
    for i, batch in enumerate(val_data):
        data = gluon.utils.split_and_load(batch[0], ctx_list=ctx, batch_axis=0)
        label = gluon.utils.split_and_load(batch[1], ctx_list=ctx, batch_axis=0)
        print(label)
        # Perform forward pass
        ort_inputs_cpu = {ort_session_cpu.get_inputs()[0].name: data[0].asnumpy()}
        print("before run")
        outputs=ort_session_cpu.run(None, ort_inputs_cpu)
        ....

When I execute the test, I obtain the following error:

multiprocessing.pool.RemoteTraceback:
"""

Traceback (most recent call last):
  File "../miniconda3/envs/ONNX_env/lib/python3.11/multiprocessing/pool.py", line 125, in worker
    result = (True, func(*args, **kwds))
                    ^^^^^^^^^^^^^^^^^^^
  File "../miniconda3/envs/ONNX_env/lib/python3.11/site-packages/mxnet/gluon/data/dataloader.py", line 400, in _worker_fn
    batch = batchify_fn([_worker_dataset[i] for i in samples])
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "../miniconda3/envs/ONNX_env/lib/python3.11/site-packages/mxnet/gluon/data/dataloader.py", line 400, in <listcomp>
    batch = batchify_fn([_worker_dataset[i] for i in samples])
                         ~~~~~~~~~~~~~~~^^^
  File "../miniconda3/envs/ONNX_env/lib/python3.11/site-packages/mxnet/gluon/data/dataset.py", line 124, in __getitem__
    return self._fn(*item)
           ^^^^^^^^^^^^^^^
  File "../TODOgooglenet_inf_get_params.py", line 51, in transform
    img = np.array(Image.fromarray(img).resize((224, 224))).astype(np.float32)
                   ^^^^^^^^^^^^^^^^^^^^
  File "../miniconda3/envs/ONNX_env/lib/python3.11/site-packages/PIL/Image.py", line 2982, in fromarray
    arr = obj.__array_interface__
          ^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: 'NDArray' object has no attribute '__array_interface__'
"""

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "../TODOgooglenet_inf_get_params.py", line 324, in <module>
    evaluate(dataset, 'B')
  File "../TODOgooglenet_inf_get_params.py", line 165, in evaluate
    for i, batch in enumerate(val_data):
  File "../miniconda3/envs/ONNX_env/lib/python3.11/site-packages/mxnet/gluon/data/dataloader.py", line 451, in __next__
    batch = pickle.loads(ret.get()) if self._dataset is None else ret.get()
                         ^^^^^^^^^
  File "../miniconda3/envs/ONNX_env/lib/python3.11/multiprocessing/pool.py", line 774, in get
    raise self._value
AttributeError: 'NDArray' object has no attribute '__array_interface__'

Any idea on this?

Start at TODOgooglenet_inf_get_params.py line 51 and inspect the object returned by ImageNet's transform before the PIL Image.fromarray call. Reproduce the failure through the DataLoader with the shown preprocessing and check whether the input type is compatible with PIL. Done means the ImageNet preprocessing completes without the AttributeError and evaluation can proceed.