Change model output to dictionary #22
Conversation
| def _run_model(self, | ||
| image, | ||
| batch_size=4, | ||
| pad_mode='constant', | ||
| preprocess_kwargs={}): | ||
| """Run the model to generate output probabilities on the data. | ||
|
|
||
| Args: | ||
| image (numpy.array): Image with shape ``[batch, x, y, channel]`` | ||
| batch_size (int): Number of images to predict on per batch. | ||
| pad_mode (str): The padding mode, one of "constant" or "reflect". | ||
| preprocess_kwargs (dict): Keyword arguments to pass to | ||
| the preprocessing function. | ||
|
|
||
| Returns: | ||
| numpy.array: Model outputs | ||
| """ | ||
| # Preprocess image if function is defined | ||
| image = self._preprocess(image, **preprocess_kwargs) | ||
|
|
||
| # Tile images, raises error if the image is not 4d | ||
| tiles, tiles_info = self._tile_input(image, pad_mode=pad_mode) | ||
|
|
||
| # Run images through model | ||
| t = timeit.default_timer() | ||
| output_tiles = self.model.predict(tiles, batch_size=batch_size) | ||
| self.logger.debug('Model inference finished in %s s', | ||
| timeit.default_timer() - t) | ||
|
|
||
| # Untile images | ||
| output_images = self._untile_output(output_tiles, tiles_info) | ||
|
|
||
| # restructure outputs into a dict if function provided | ||
| formatted_images = {name: pred for name, pred in zip(self.model.output_names, | ||
| output_images)} | ||
|
|
||
| return formatted_images | ||
|
|
There was a problem hiding this comment.
Why did you decide to duplicate _run_model with the addition of lines 142-144 instead of just adding those lines after calling _run_model in _predict?
There was a problem hiding this comment.
I replaced the line calling _format_model_output because it didn't have a way to pass arguments in the _format_model_output_fn. You're right that I could have just done this outside of _run_model but my thought was that the line I added serves the same purpose as the _format_model_output call.
There was a problem hiding this comment.
I would try to avoid duplicating a core function like _run_model especially if there is a way to pass in a format function. It's going to make the spots repo harder to maintain in the long run if we ever do a major refactor of the deepcell application code. If you want keep the output names exposed, you could make it a parameter in a function format_spots_output and then when you pass it in to the application to define _format_model_output_fn you could first override it with self.model.output_names, e.g.
def _format_spots_output(output):
return format_spots_otuput(output, output_names=self.model.output_names)
I'll let it go if you think its too complicated, but take a look first.
There was a problem hiding this comment.
If you think it's cleaner to move it out of _run_model I can change it
There was a problem hiding this comment.
I would prefer having that change.
This PR makes the changes necessary to name the outputs of the
dot_net_2Dmodel after prediction in a dictionary as opposed to a list. I have also removed arguments for scaling from the application, which is not currently supported by the spots model.