Skip to content

Fixes #8697 GPU memory leak by checking both image and label tensors for CUDA device #8708

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
benediktjohannes wants to merge 9 commits into
base: dev
Choose a base branch
from
Open

Fixes #8697 GPU memory leak by checking both image and label tensors for CUDA device #8708

Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
aa93e6c
Fixes #8697 GPU memory leak by checking both image and label tensors …
benediktjohannes Jan 17, 2026
f8cc9ff
Forgot to define variables, thanks coderabbitai for the information
benediktjohannes Jan 17, 2026
c532535
Update analyzer.py
benediktjohannes Jan 24, 2026
4367d32
Merge branch 'Project-MONAI:dev' into patch-1
benediktjohannes Feb 15, 2026
546487d
Merge branch 'dev' into patch-1
ericspod Feb 20, 2026
a83b4bb
Merge branch 'dev' into patch-1
ericspod Feb 21, 2026
28539e2
Merge branch 'Project-MONAI:dev' into patch-1
benediktjohannes Feb 22, 2026
a963ac8
Update analyzer.py
benediktjohannes Feb 22, 2026
2820618
Update analyzer.py
benediktjohannes 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
16 changes: 9 additions & 7 deletions monai/auto3dseg/analyzer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -468,21 +468,23 @@ def __call__(self, data: Mapping[Hashable, MetaTensor]) -> dict[Hashable, MetaTe
"""
d: dict[Hashable, MetaTensor] = dict(data)
start = time.time()
if isinstance(d[self.image_key], (torch.Tensor, MetaTensor)) and d[self.image_key].device.type == "cuda":
using_cuda = True
else:
using_cuda = False
image_tensor = d[self.image_key]
label_tensor = d[self.label_key]
using_cuda = any(
isinstance(t, (torch.Tensor, MetaTensor)) and t.device.type == "cuda"
for t in (image_tensor, label_tensor)
)
restore_grad_state = torch.is_grad_enabled()
torch.set_grad_enabled(False)

ndas: list[MetaTensor] = [d[self.image_key][i] for i in range(d[self.image_key].shape[0])] # type: ignore
ndas_label: MetaTensor = d[self.label_key].astype(torch.int16) # (H,W,D)
ndas: list[MetaTensor] = [image_tensor[i] for i in range(image_tensor.shape[0])] # type: ignore
ndas_label: MetaTensor = label_tensor.astype(torch.int16) # (H,W,D)
Comment on lines 471 to 481
Copy link
Contributor

@coderabbitai coderabbitai bot Jan 24, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

⚠️ Potential issue | 🟡 Minor

🧩 Analysis chain

🏁 Script executed:

rg -n "LabelStats|FgImageStats|using_cuda|empty_cache|get_label_ccp" -g "*test*.py" -i

Repository: Project-MONAI/MONAI

Length of output: 1005

🏁 Script executed:

# Also check the actual code at the specified lines
head -500 monai/auto3dseg/analyzer.py | tail -50

Repository: Project-MONAI/MONAI

Length of output: 2358

🏁 Script executed:

# Find the LabelStats class and its __call__ method
rg -n "class LabelStats|def __call__" monai/auto3dseg/analyzer.py

Repository: Project-MONAI/MONAI

Length of output: 662

🏁 Script executed:

sed -n '319,360p' tests/apps/test_auto3dseg.py

Repository: Project-MONAI/MONAI

Length of output: 2553

🏁 Script executed:

sed -n '341,370p' tests/apps/test_auto3dseg.py

Repository: Project-MONAI/MONAI

Length of output: 1890

🏁 Script executed:

sed -n '420,500p' tests/apps/test_auto3dseg.py

Repository: Project-MONAI/MONAI

Length of output: 3971

🏁 Script executed:

head -50 tests/apps/test_auto3dseg.py | grep -E "device|import"

Repository: Project-MONAI/MONAI

Length of output: 606

🏁 Script executed:

grep -n "^device\s*=" tests/apps/test_auto3dseg.py

Repository: Project-MONAI/MONAI

Length of output: 80

🏁 Script executed:

# Check the beginning of the test file for device setup
head -100 tests/apps/test_auto3dseg.py

Repository: Project-MONAI/MONAI

Length of output: 3483

🏁 Script executed:

grep -n "SIM_GPU_TEST_CASES\|@parameterized\|skip_if_no_cuda" tests/apps/test_auto3dseg.py

Repository: Project-MONAI/MONAI

Length of output: 359

🏁 Script executed:

# Check if there are GPU parametrized test methods
grep -B 5 -A 10 "test_label_stats.*gpu\|test_label_stats.*cuda" tests/apps/test_auto3dseg.py

Repository: Project-MONAI/MONAI

Length of output: 45

🏁 Script executed:

sed -n '214,280p' tests/apps/test_auto3dseg.py

Repository: Project-MONAI/MONAI

Length of output: 3116

🏁 Script executed:

# Check what the GPU test actually tests
grep -A 40 "@skip_if_no_cuda" tests/apps/test_auto3dseg.py | head -60

Repository: Project-MONAI/MONAI

Length of output: 1974

Add tests for mixed-device LabelStats handling.

Tests currently cover CPU-only and GPU-only scenarios but not mixed (image on CPU + label on CUDA, or vice versa). Add parameterized test cases to verify LabelStats handles these mixed-device scenarios correctly per the coding guidelines.

🤖 Prompt for AI Agents 
In `@monai/auto3dseg/analyzer.py` around lines 471 - 478, Add parameterized unit
tests that cover mixed-device scenarios for LabelStats by creating cases where
image_tensor is on CPU and label_tensor is on CUDA and vice versa; instantiate
batches similar to how ndas and ndas_label are derived (use image_tensor,
label_tensor and MetaTensor/torch.Tensor on specific torch.device settings),
call the LabelStats code paths that consume ndas/ndas_label, and assert expected
statistics and that no device-related errors occur. Ensure tests toggle CUDA
availability with torch.cuda.is_available() guards, use explicit .to(device) on
tensors, and verify behavior matches the existing CPU-only and GPU-only
assertions so mixed-device handling is validated.


if ndas_label.shape != ndas[0].shape:
raise ValueError(f"Label shape {ndas_label.shape} is different from image shape {ndas[0].shape}")

nda_foregrounds: list[torch.Tensor] = [get_foreground_label(nda, ndas_label) for nda in ndas]
nda_foregrounds = [nda if nda.numel() > 0 else torch.Tensor([0]) for nda in nda_foregrounds]
nda_foregrounds = [nda if nda.numel() > 0 else MetaTensor([0.0]) for nda in nda_foregrounds]

unique_label = unique(ndas_label)
if isinstance(ndas_label, (MetaTensor, torch.Tensor)):
Expand Down
Loading