Understanding of spatial feature map

[czhang0528]

text_objseg/models/processing_tools.py

Line 6 in be664d6

spatial_batch_val = np.zeros((N, featmap_H, featmap_W, 8), dtype=np.float32)

Hi, may I ask why spatial channel number is eight rather than two, which mismatches the claim in the paper:

...we obtain a w×h×(D_im +2) representation containing local image descriptors and spatial coordinates...

Could you help to clarify it or am I missing anything else? Thank you.