did VotePooling implement right?

[world4jason]

original paper set k = 7
k=3 is for visualization purpose
so a ROI block crop(ROI_ALIGN) from score_map might have the following dimension
[ 7, 7, n_classes ]

but the code seems weird here?
should the code change from
pooled.append(tf.image.crop_and_resize( score_maps[i], level_boxes, box_indices, [self.pool_shape * self.k, self.pool_shape * self.k], method="bilinear"))

to something like this?

pooled.append(tf.image.crop_and_resize( score_maps[i], level_boxes, box_indices, [ self.k, self.k], method="bilinear"))

[world4jason]

I'm really confused that why here is a K and also a pool_shape?

btw, in light-head rcnn also take k=7 as an example when they plot the comparison picture in figure 2
https://arxiv.org/pdf/1711.07264.pdf

[kevinn888]

i also got an error in votePooling when i try to run this code, did you already know the answer..??