Skip to content

Refactor label handling in cross_entropy_loss.py #12431

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
shyhyawJou wants to merge 2 commits into
base: main
Choose a base branch
from
+14 −0
Open

Refactor label handling in cross_entropy_loss.py #12431

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
3b1174b
Refactor label handling in cross_entropy_loss.py
shyhyawJou Feb 22, 2026
4b9c9ef
Clean up whitespace in cross_entropy_loss.py
shyhyawJou Feb 22, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 14 additions & 0 deletions mmdet/models/losses/cross_entropy_loss.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,6 +39,11 @@ def cross_entropy(pred,
"""
# The default value of ignore_index is the same as F.cross_entropy
ignore_index = -100 if ignore_index is None else ignore_index

# label[0] are categorical labels of images
if isinstance(label, tuple) and len(label) > 1:
label = label[0]

# element-wise losses
loss = F.cross_entropy(
pred,
Expand Down Expand Up @@ -117,6 +122,10 @@ def binary_cross_entropy(pred,
# The default value of ignore_index is the same as F.cross_entropy
ignore_index = -100 if ignore_index is None else ignore_index

# label[0] are categorical labels of images
if isinstance(label, tuple) and len(label) > 1:
label = label[0]

if pred.dim() != label.dim():
label, weight, valid_mask = _expand_onehot_labels(
label, weight, pred.size(-1), ignore_index)
Expand Down Expand Up @@ -191,6 +200,11 @@ def mask_cross_entropy(pred,
assert ignore_index is None, 'BCE loss does not support ignore_index'
# TODO: handle these two reserved arguments
assert reduction == 'mean' and avg_factor is None

# label[0] are categorical labels of images
if isinstance(label, tuple) and len(label) > 1:
label = label[0]

num_rois = pred.size()[0]
inds = torch.arange(0, num_rois, dtype=torch.long, device=pred.device)
pred_slice = pred[inds, label].squeeze(1)
Expand Down