Describe the issue:
When settings["registration"]["block_size"] is set to (Ly, x) or (y, Lx), non-rigid registration fails.
I believe you can fix this by replacing code in nonrigid.transform_data with:
# Ensure interpolate output size is valid even for a single block in a dimension
Lyc = max(1, int(yb.max() - yb.min()))
Lxc = max(1, int(xb.max() - xb.min()))
yxup = F.interpolate(
torch.stack((ymax1, xmax1), dim=1),
size=(Lyc, Lxc),
mode="bilinear",
align_corners=True,
)
# If you pad later, make the single-block case safe (right/bottom must be non-negative)
xb_min = int(xb.min())
xb_max = int(xb.max())
yb_min = int(yb.min())
yb_max = int(yb.max())
pad_left = max(0, xb_min)
pad_right = max(0, Lx - xb_max)
pad_top = max(0, yb_min)
pad_bottom = max(0, Ly - yb_max)
yxup = F.pad(yxup, (pad_left, pad_right, pad_top, pad_bottom), mode="replicate")
Reproduce the code example:
pseudo-code:
from suite2p.registration import register
settings["registration"]["block_size"] = [64, 256]
# f_reg.shape: [500, 256, 256]
register.registration_wrapper(f_reg, settings=settings)
Example input dataset used to reproduce the issue:
No response
Error message:
File "C:\dev\workspaces\SchroederLab\depth-for-2p\.venv\Lib\site-packages\suite2p\registration\register.py", line 933, in registration_wrapper
outputs = register_frames(f_align_in, f_align_out=f_align_out, bidiphase=bidiphase,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\dev\workspaces\SchroederLab\depth-for-2p\.venv\Lib\site-packages\suite2p\registration\register.py", line 609, in register_frames
frames = shift_frames(fr_torch, ymax, xmax, ymax1, xmax1, blocks,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\dev\workspaces\SchroederLab\depth-for-2p\.venv\Lib\site-packages\suite2p\registration\register.py", line 447, in shift_frames
fr_torch = nonrigid.transform_data(fr_torch, blocks[2], blocks[1], blocks[0],
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\dev\workspaces\SchroederLab\depth-for-2p\.venv\Lib\site-packages\suite2p\registration\nonrigid.py", line 414, in transform_data
yxup = F.interpolate(torch.stack((ymax1, xmax1), dim=1),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\dev\workspaces\SchroederLab\depth-for-2p\.venv\Lib\site-packages\torch\nn\functional.py", line 4867, in interpolate
return torch._C._nn.upsample_bilinear2d(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Input and output sizes should be greater than 0, but got input (H: 6, W: 1) output (H: 192, W: 0)Version information:
suite2p v1.0.0.2.dev11+g20ea7e15e
Context for the issue:
No response
Describe the issue:
When settings["registration"]["block_size"] is set to (Ly, x) or (y, Lx), non-rigid registration fails.
I believe you can fix this by replacing code in nonrigid.transform_data with:
Reproduce the code example:
Example input dataset used to reproduce the issue:
No response
Error message:
Version information:
suite2p v1.0.0.2.dev11+g20ea7e15e
Context for the issue:
No response