def __call__(self, match_quality_matrix, batched = 0):
"""
Args:
match_quality_matrix (Tensor[float]): an MxN tensor, containing the
pairwise quality between M ground-truth elements and N predicted elements.
Returns:
matches (Tensor[int64]): an N tensor where N[i] is a matched gt in
[0, M - 1] or a negative value indicating that prediction i could not
be matched.
"""
if match_quality_matrix.numel() == 0:
# empty targets or proposals not supported during training
if match_quality_matrix.shape[0] == 0:
raise ValueError(
"No ground-truth boxes available for one of the images "
"during training")
else:
raise ValueError(
"No proposal boxes available for one of the images "
"during training")
# match_quality_matrix is M (gt) x N (predicted)
# Max over gt elements (dim 0) to find best gt candidate for each prediction
if match_quality_matrix.is_cuda:
if batched:
matches = _C.match_proposals(match_quality_matrix,self.allow_low_quality_matches, self.low_threshold, self.high_threshold)
else:
match_quality_matrix_unsqueezed = match_quality_matrix.unsqueeze(0)
matches = _C.match_proposals(match_quality_matrix_unsqueezed, self.allow_low_quality_matches, self.low_threshold, self.high_threshold).squeeze(0)
else:
matched_vals, matches = match_quality_matrix.max(dim=0)
if self.allow_low_quality_matches:
all_matches = matches.clone()
# Assign candidate matches with low quality to negative (unassigned) values
below_low_threshold = matched_vals < self.low_threshold
between_thresholds = (matched_vals >= self.low_threshold) & (
matched_vals < self.high_threshold
)
matches.masked_fill_(below_low_threshold, Matcher.BELOW_LOW_THRESHOLD)
matches.masked_fill_(between_thresholds, Matcher.BETWEEN_THRESHOLDS)
if self.allow_low_quality_matches:
self.set_low_quality_matches_(matches, all_matches, match_quality_matrix)
return matches
def __call__(self, match_quality_matrix):
"""
Args:
match_quality_matrix (Tensor[float]): an MxN tensor, containing the
pairwise quality between M ground-truth elements and N predicted elements.
Returns:
matches (Tensor[float]): an N tensor where N[i] is a matched gt in
[0, M - 1] or a negative value indicating that prediction i could not
be matched.
"""
# match_quality_matrix is M (gt) x N (predicted)
matches = _C.match_proposals(match_quality_matrix,
self.allow_low_quality_matches,
self.low_threshold, self.high_threshold)
return matches