Transfer learning y convolucional #2
Conversation
Change convolutional architecture
… base class. Implement transfer learning
# Conflicts: # notebooks/model_training.ipynb # src/catdog/catdog/dataset.py # src/catdog/catdog/models/base.py
src/catdog/catdog/dataset.py
Outdated
|
|
||
| from definitions import IMG_PATH | ||
|
|
||
|
|
||
|
|
||
| class CatDogDataset(Dataset): | ||
| def __init__(self, cat_dog_df, transforms=None, img_output_size=(500,500)): | ||
| def __init__(self, cat_dog_df, transformations=None, feature_scaling=255, img_output_size=(500, 500), img_path="../data/images/"): |
There was a problem hiding this comment.
img_path hardcodeado lo queres ahi?? yo lo habia reemplazado por definitions.IMG_PATH
There was a problem hiding this comment.
No, debí mirar con poco amor al hacer el merge del archivo con main. Corrijo en próximo commit.
| self.resizer = transforms.Resize(img_output_size) | ||
|
|
||
| self.to_tensor = transforms.ToTensor() | ||
| self.normalizer = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) |
There was a problem hiding this comment.
esto no es hardcode??
There was a problem hiding this comment.
ahi vi que se usa eso como estandar de pytorch. lo queremos usar para todos nuestros modelos, aun los que no fueron pre entrenados??
pytorch/vision#1439
There was a problem hiding this comment.
Exactamente por eso lo puse. Osea sí, mucho no me lo cuestioné. Podemos calcular esos valores con nuestro dataset y reemplazarlos, pero no me calienta mucho diría.
src/catdog/catdog/dataset.py
Outdated
| if self.transforms is not None: | ||
| np_img, bbox = self.transforms(np_img, bbox) | ||
| if self.transformations is not None: | ||
| np_img, bbox = self.transformations(img, bbox) |
There was a problem hiding this comment.
aca tal vez queremos renombrar np_img??
There was a problem hiding this comment.
De hecho sí, es un bug. Corrijo en próximo commit.
No description provided.