Skip to content
New issue

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

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

Already on GitHub? # to your account

[Feature] Add mask channel in MGD Loss #461

Merged
merged 3 commits into from
Mar 1, 2023
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
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
8 changes: 7 additions & 1 deletion mmrazor/models/architectures/connectors/mgd_connector.py
Original file line number Diff line number Diff line change
Expand Up @@ -27,10 +27,12 @@ def __init__(
student_channels: int,
teacher_channels: int,
lambda_mgd: float = 0.65,
mask_on_channel: bool = False,
init_cfg: Optional[Dict] = None,
) -> None:
super().__init__(init_cfg)
self.lambda_mgd = lambda_mgd
self.mask_on_channel = mask_on_channel
if student_channels != teacher_channels:
self.align = nn.Conv2d(
student_channels,
Expand All @@ -55,7 +57,11 @@ def forward_train(self, feature: torch.Tensor) -> torch.Tensor:
N, C, H, W = feature.shape

device = feature.device
mat = torch.rand((N, 1, H, W)).to(device)
if not self.mask_on_channel:
mat = torch.rand((N, 1, H, W)).to(device)
else:
mat = torch.rand((N, C, 1, 1)).to(device)

mat = torch.where(mat > 1 - self.lambda_mgd,
torch.zeros(1).to(device),
torch.ones(1).to(device)).to(device)
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -144,6 +144,22 @@ def test_mgd_connector(self):
assert s_output1.shape == torch.Size([1, 16, 8, 8])
assert s_output2.shape == torch.Size([1, 32, 8, 8])

mgd_connector1 = MGDConnector(
student_channels=16,
teacher_channels=16,
lambda_mgd=0.65,
mask_on_channel=True)
mgd_connector2 = MGDConnector(
student_channels=16,
teacher_channels=32,
lambda_mgd=0.65,
mask_on_channel=True)
s_output1 = mgd_connector1.forward_train(s_feat)
s_output2 = mgd_connector2.forward_train(s_feat)

assert s_output1.shape == torch.Size([1, 16, 8, 8])
assert s_output2.shape == torch.Size([1, 32, 8, 8])

def test_norm_connector(self):
s_feat = torch.randn(2, 3, 2, 2)
norm_cfg = dict(type='BN', affine=False, track_running_stats=False)
Expand Down