Detection model #32
Detection model #32
Conversation
| @@ -1,5 +1,4 @@ | |||
| # Copyright (C) 2021, Mindee. | |||
|
|
|||
There was a problem hiding this comment.
Better to keep this line break
|
|
||
| class DetectionModel(keras.Model): | ||
| """Implements abstract DetectionModel class | ||
|
|
doctr/models/detection/model.py
Outdated
| self, | ||
| inputs: tf.Tensor, | ||
| training: bool = False | ||
| ) -> Tuple[keras.Model, keras.Model]: |
| @@ -161,3 +164,248 @@ def __call__( | |||
|
|
|||
| bounding_boxes.append(boxes_batch) | |||
| return bounding_boxes | |||
|
|
|||
|
|
|||
| class DBModel(DetectionModel, keras.Model): | |||
There was a problem hiding this comment.
The inherited DetectionModel already inherits from keras.Model
Also, since we will use ResNet50 for now, let's name it DBResNet50
test/test_models.py
Outdated
| assert np.shape(dboutput_notrain.numpy())[0] == 8 | ||
| # output dimensions | ||
| assert np.shape(dboutput_notrain.numpy())[1] == np.shape(dboutput_notrain.numpy())[2] == 640 |
There was a problem hiding this comment.
you can regroup shape value testing:
assert dboutput_notrain.numpy().shape == (8, 640, 640, 3)
test/test_models.py
Outdated
| assert all(np.shape(maps.numpy())[0] == 8 for maps in dboutput_train) | ||
| # output dimensions | ||
| assert all(np.shape(maps.numpy())[1] == np.shape(maps.numpy())[2] == 640 for maps in dboutput_train) |
There was a problem hiding this comment.
Same here, this can be grouped (and np.shape is not required, you can use the .shape method of the ndarray)
| channels: int = 128, | ||
| ) -> None: | ||
| super().__init__(shape) | ||
| self.channels = channels |
There was a problem hiding this comment.
Many modules could be instantiated over here
Codecov Report
@@ Coverage Diff @@
## main #32 +/- ##
=======================================
Coverage 98.43% 98.43%
=======================================
Files 12 13 +1
Lines 256 320 +64
=======================================
+ Hits 252 315 +63
- Misses 4 5 +1
Continue to review full report at Codecov.
|
DetectionModel class and DBModel (child class) added.
DB is implemented with a ResNet50 backbone (feature extactor) provided by keras