the VP part problem about vpgnet HOT 4 OPEN

layer {
  name: "vp-conv"
  type: "Convolution"
  bottom: "L6d"
  top: "vp-conv"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}

layer {
  name: "vp-tile"
  type: "Tiling"
  bottom: "vp-conv"
  top: "vp-conv-tiled"
  tiling_param {
    tile_dim: 8
  }
}

# Pixel level softmax prediction.
layer {
  name: "vp-prob"
  type: "Softmax"
  bottom: "vp-conv-tiled"
  top: "vp-prob"
}

vp_prob = net.blobs['vp-prob'].data[0]
[vp_y, vp_x] = vp_quadrant(vp_prob)

def vp_quadrant(prob):
	prob[0,:,:] = scipy.ndimage.filters.gaussian_filter(prob[0,:,:], 5)
	prob[1,:,:] = scipy.ndimage.filters.gaussian_filter(prob[1,:,:], 5)
	prob[2,:,:] = scipy.ndimage.filters.gaussian_filter(prob[2,:,:], 5)
	prob[3,:,:] = scipy.ndimage.filters.gaussian_filter(prob[3,:,:], 5)
	prob[4,:,:] = scipy.ndimage.filters.gaussian_filter(prob[4,:,:], 5)

	P_avg = (1 - np.mean(prob[0,:,:])) / 4
	avg_map = np.ones(prob[0,:,:].shape) * P_avg
	loc_map = np.power((avg_map-prob[1,:,:]), 2) + np.power((avg_map-prob[2,:,:]), 2) + np.power((avg_map-prob[3,:,:]), 2) + np.power((avg_map-prob[4,:,:]), 2)

	[y, x] = np.unravel_index(loc_map.argmin(), loc_map.shape)
	return [y, x]