Place205 VS Place365 mean pixel subtraction

[mrgransky]

In Place205, input images are zero-centered by mean pixel (rather than mean image) subtraction with the following values explained in readme.txt file when downloaded here:

mu = np.array([105.487823486, 113.741088867, 116.060394287])

The following function with places365CNN_mean.binaryproto:

def load_mean_file(mean_file):
        if mean_file.endswith('.npy'):
	    return np.load(mean_file)
        with open(mean_file, 'rb') as infile:
            blob = caffe_pb2.BlobProto()
            blob.MergeFromString(infile.read())
            if blob.HasField('shape'):
                blob_dims = blob.shape
                assert len(blob_dims) == 4, 'Shape should have 4 dimensions - shape is "%s"' % blob.shape
            elif blob.HasField('num') and blob.HasField('channels') and \
                    blob.HasField('height') and blob.HasField('width'):
                blob_dims = (blob.num, blob.channels, blob.height, blob.width)
            else:
                raise ValueError('blob does not provide shape or 4d dimensions')
            mean=np.reshape(blob.data, blob_dims[1:])
        return mean 

returns: mu = np.array([104.05100722, 112.51448911, 116.67603893])

I was just wondering if this mean value for Place365 dataset seems correct?

Cheers,