Unet integration #44
Unet integration #44
Conversation
…ion with only one mask
There was a problem hiding this comment.
Hi @KorenMary, please take a look at my comments, some small changes are needed but all in all it is looking good thanks :)
There was a problem hiding this comment.
can you remove this? you can add your configs to your gitignore so we you don't accidentaly commit changes to these files when playing around withe the configurations
There was a problem hiding this comment.
are these changes part of the work for the front-end extension to work with multiple masks?
There was a problem hiding this comment.
rename unet to end-2-end
| @@ -45,8 +48,10 @@ def eval(self, img): | |||
| :return: mask of the image, list of 2D arrays, or single 3D array (if do_3D=True) labelled image. | |||
| :rtype: np.ndarray | |||
| """ | |||
| return super().eval(x=img, **self.eval_config["segmentor"])[0] # 0 to take only mask | |||
There was a problem hiding this comment.
are changes to CustomCellposeModel necessary for Unet integration?
| loss.backward() | ||
| optimizer.step() | ||
|
|
||
| self.loss /= len(train_dataloader) |
There was a problem hiding this comment.
you are not adding something to self.loss, only computing loss in the loop
| img = img.unsqueeze(1) if img.ndim == 3 else img | ||
| # convert to tensor | ||
| # img = torch.permute(torch.tensor(img.astype(np.float32)), (2, 0, 1)).unsqueeze(0) | ||
|
|
There was a problem hiding this comment.
add model.eval or torch.no_grad?
|
|
||
| y_hat = np.stack((y_instances, y_labels)) | ||
|
|
||
| return y_hat |
There was a problem hiding this comment.
can you rename these variables (y_hat, y_instances etc.) to the variable names we used in CellPosePatchCNN, so we are consistent in our naming?
No description provided.