ALEXNET: "ValueError: string_input_producer requires a non-null input tensor"

[AntonioSerrano]

I am trying to run the benchmark for Alexnet via ./fathom/alexnet/alexnet.py. But I am getting the following error:

Traceback (most recent call last):
  File "./fathom/alexnet/alexnet.py", line 137, in <module>
    m = AlexNet()
  File "/usr/local/lib/python2.7/dist-packages/Fathom_Workloads-1.0rc0-py2.7.egg/fathom/nn.py", line 45, in __init__
    self.build()
  File "/usr/local/lib/python2.7/dist-packages/Fathom_Workloads-1.0rc0-py2.7.egg/fathom/nn.py", line 105, in build
    self.build_inputs()
  File "/usr/local/lib/python2.7/dist-packages/Fathom_Workloads-1.0rc0-py2.7.egg/fathom/imagenet/imagenet.py", line 59, in build_inputs
    self.batch_images_queue, self.batch_labels_queue = distorted_inputs(self.dataset, batch_size=self.batch_size)
  File "/usr/local/lib/python2.7/dist-packages/Fathom_Workloads-1.0rc0-py2.7.egg/fathom/imagenet/image_processing.py", line 133, in distorted_inputs
    num_preprocess_threads=num_preprocess_threads)
  File "/usr/local/lib/python2.7/dist-packages/Fathom_Workloads-1.0rc0-py2.7.egg/fathom/imagenet/image_processing.py", line 425, in batch_inputs
    filename_queue = tf.train.string_input_producer(data_files, capacity=16)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/training/input.py", line 217, in string_input_producer
    raise ValueError(not_null_err)
ValueError: string_input_producer requires a non-null input tensor

It seems like TensorFlow in image_processing.py is trying to use the images located in /data/ILSVRC2012/imagenet-tfrecord but it is not able to find them or they are not formatted as expected. I downloaded the 138 Gb from ILSVRC2012 and put them in the mentioned path (/data/ILSVRC2012/imagenet-tfrecord). But the truth is that it does not work.

Any clue about what I am doing wrong? Many thx.