def crop_and_resize(self,
bboxes,
out_shape,
inds,
device='cpu',
interpolation='bilinear',
binarize=True):
"""See :func:`BaseInstanceMasks.crop_and_resize`."""
if len(self.masks) == 0:
empty_masks = np.empty((0, *out_shape), dtype=np.uint8)
return BitmapMasks(empty_masks, *out_shape)
# convert bboxes to tensor
if isinstance(bboxes, np.ndarray):
bboxes = torch.from_numpy(bboxes).to(device=device)
if isinstance(inds, np.ndarray):
inds = torch.from_numpy(inds).to(device=device)
num_bbox = bboxes.shape[0]
fake_inds = torch.arange(
num_bbox, device=device).to(dtype=bboxes.dtype)[:, None]
rois = torch.cat([fake_inds, bboxes], dim=1) # Nx5
rois = rois.to(device=device)
if num_bbox > 0:
gt_masks_th = torch.from_numpy(self.masks).to(device).index_select(
0, inds).to(dtype=rois.dtype)
targets = roi_align(gt_masks_th[:, None, :, :], rois, out_shape,
1.0, 0, 'avg', True).squeeze(1)
if binarize:
resized_masks = (targets >= 0.5).cpu().numpy()
else:
resized_masks = targets.cpu().numpy()
else:
resized_masks = []
return BitmapMasks(resized_masks, *out_shape)
def forward(self, instance_feats, semantic_feat, semantic_pred, rois,
roi_labels):
concat_tensors = [instance_feats]
# instance-wise semantic feats
semantic_feat = self.relu(self.semantic_transform_in(semantic_feat))
ins_semantic_feats = self.semantic_roi_extractor([
semantic_feat,
], rois)
ins_semantic_feats = self.relu(
self.semantic_transform_out(ins_semantic_feats))
concat_tensors.append(ins_semantic_feats)
# instance masks
instance_preds = self.instance_logits(instance_feats)[
torch.arange(len(rois)), roi_labels][:, None]
_instance_preds = instance_preds.sigmoid(
) if self.mask_use_sigmoid else instance_preds
instance_masks = F.interpolate(_instance_preds,
instance_feats.shape[-2],
mode='bilinear',
align_corners=True)
concat_tensors.append(instance_masks)
# instance-wise semantic masks
fake_rois = rois.clone()
fake_rois[:, 0] = torch.zeros(len(rois))
_semantic_pred = semantic_pred.sigmoid(
) if self.mask_use_sigmoid else semantic_pred
ins_semantic_masks = roi_align(_semantic_pred, fake_rois,
instance_feats.shape[-2:],
1.0 / self.semantic_out_stride, 0,
'avg', True)
ins_semantic_masks = F.interpolate(ins_semantic_masks,
instance_feats.shape[-2:],
mode='bilinear',
align_corners=True)
concat_tensors.append(ins_semantic_masks)
# fuse instance feats & instance masks & semantic feats & semantic masks
fused_feats = torch.cat(concat_tensors, dim=1)
for conv in self.fuse_conv:
fused_feats = self.relu(conv(fused_feats))
fused_feats = self.relu(self.fuse_transform_out(fused_feats))
fused_feats = self.relu(self.upsample(fused_feats))
# concat instance and semantic masks with fused feats again
instance_masks = F.interpolate(_instance_preds,
fused_feats.shape[-2],
mode='bilinear',
align_corners=True)
ins_semantic_masks = F.interpolate(ins_semantic_masks,
fused_feats.shape[-2],
mode='bilinear',
align_corners=True)
fused_feats = torch.cat(
[fused_feats, instance_masks, ins_semantic_masks], dim=1)
return instance_preds, fused_feats