Skip to content

Improvements to the training UX #2775

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

Merged
n-poulsen merged 7 commits into from
Nov 8, 2024
Merged

Improvements to the training UX #2775

Changes from 1 commit
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
751fd37
minor training improvements
n-poulsen Nov 6, 2024
bc38e1d
fix detector config creation to include best metric. fixed snapshot m…
n-poulsen Nov 6, 2024
810fc1a
add option to resume training
n-poulsen Nov 6, 2024
525b72d
round with numpy as torch.round is not implemented with MPS
n-poulsen Nov 6, 2024
d7401ff
fix state dict saving when training on multiple GPUs
n-poulsen Nov 6, 2024
d152ca1
bug fix: only print metrics at epochs where eval occured
n-poulsen Nov 6, 2024
7ec0d17
typo fixes
n-poulsen Nov 8, 2024
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
Prev Previous commit
Next Next commit
round with numpy as torch.round is not implemented with MPS
  • Loading branch information
@n-poulsen
n-poulsen committed Nov 6, 2024
commit 525b72df45f39109694effa67b91b7f719085b07
7 changes: 3 additions & 4 deletions deeplabcut/pose_estimation_pytorch/models/predictors/paf_predictor.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -212,14 +212,13 @@ def calc_peak_locations(
locrefs: torch.Tensor,
peak_inds_in_batch: torch.Tensor,
strides: tuple[float, float],
n_decimals: int = 3,
) -> torch.Tensor:
s, b, r, c = peak_inds_in_batch.T
stride_y, stride_x = strides
strides = torch.Tensor((stride_x, stride_y)).to(locrefs.device)
off = locrefs[s, b, :, r, c]
loc = strides * peak_inds_in_batch[:, [3, 2]] + strides // 2 + off
return torch.round(loc, decimals=n_decimals)
return loc

@staticmethod
def compute_edge_costs(
Expand Down Expand Up @@ -331,8 +330,8 @@ def compute_peaks_and_costs(
) -> list[dict[str, NDArray]]:
n_samples, n_channels = heatmaps.shape[:2]
n_bodyparts = n_channels - n_id_channels
pos = self.calc_peak_locations(locrefs, peak_inds_in_batch, strides, n_decimals)
pos = pos.detach().cpu().numpy()
pos = self.calc_peak_locations(locrefs, peak_inds_in_batch, strides)
pos = np.round(pos.detach().cpu().numpy(), decimals=n_decimals)
heatmaps = heatmaps.detach().cpu().numpy()
pafs = pafs.detach().cpu().numpy()
peak_inds_in_batch = peak_inds_in_batch.detach().cpu().numpy()
Expand Down