def compute_conditional_samples(batch, *args):
image_batch = batch.cuda()
_ = model(image_batch)
acts = model.retained_layer(layername)
seg = segmodel.segment_batch(renorm(image_batch), downsample=4)
hacts = upfn(acts)
return tally.conditional_samples(hacts, seg)
def compute_conditional_indicator(batch, *args):
image_batch = batch.cuda()
seg = segmodel.segment_batch(renorm(image_batch), downsample=4)
_ = model(image_batch)
acts = model.retained_layer(layername)
hacts = upfn(acts)
iacts = (hacts > level_at_99).float() # indicator where > 0.99 percentile.
return tally.conditional_samples(iacts, seg)
def compute_conditional_indicator(batch, *args):
data_batch = batch.cuda()
out_batch = model(data_batch)
image_batch = out_batch if is_generator else renorm(data_batch)
seg = segmodel.segment_batch(image_batch, downsample=4)
acts = model.retained_layer(layername)
hacts = upfn(acts)
iacts = (hacts > level_at_99).float() # indicator
return tally.conditional_samples(iacts, seg)
def compute_selected_segments(batch, *args):
img, seg = batch
# show(iv.segmentation(seg))
image_batch = img.cuda()
seg_batch = seg.cuda()
_ = self.model(image_batch)
acts = self.model.retained_layer(self.layername)
hacts = self.upfn(acts)
iacts = (hacts >
level_at_99).float() # indicator where > 0.99 percentile.
return tally.conditional_samples(iacts, seg_batch)