Extension type improvements to support 3d types #173
Conversation
| raise ValueError("ndims must be equal to compute IOU") | ||
| size_self = self._box_sizes() | ||
| if self is others: | ||
| size_others = size_self |
There was a problem hiding this comment.
should we just return np.array([1, 1, 1, 1, ...]) here?
Otherwise, this seems to just save one compute of box sizes? i feel that it is not necessary at the moment.
There was a problem hiding this comment.
it's all-pairs so even if we're doing box_arr.iou(box_arr), it should return a matrix of ious no?
There was a problem hiding this comment.
then it returns a np.ones(shape=(self.array.length, self.array.length))?
There was a problem hiding this comment.
Why would it return all ones? If you have a list of boxes, box_i.iou(box_j) won't always be 1?
| ) | ||
|
|
||
| def get_length(self, axis: Union[int, str]): |
There was a problem hiding this comment.
Shall we make a new name of it?
i.e, a 2d box (rectangle) has width / length , but this get_length() is actually capable to get both IIUC.
There was a problem hiding this comment.
good point. This is the edge length along a given dimension.
how about get_edge_len ?
| return self.get_max("y") | ||
|
|
||
| def _box_sizes(self): | ||
| sizes = self.get_length(0) |
There was a problem hiding this comment.
shall we do
sizes = np.ones_like(self.dims[0])
for i in range(self.ndims):
sizes *= self.get_length(i)
Or prob a reduce .
There was a problem hiding this comment.
wanted to save the allocation and one array mult operation. i guess reduce could work well.
There was a problem hiding this comment.
changed to using np.multiply.reduce
changhiskhan
force-pushed
the
changhiskhan/extensions-3d
branch
from
4335361 to
5db9ef5
Compare
Support point2d, box 2d/3d
Also added polygon 2d/3d to support segmentation