Hannes Kuchelmeister
commited on
Commit
·
a490849
1
Parent(s):
c3f6f40
add mean laplacian and mdct
Browse files
notebooks/5.0-hfk-comparing-to-traditional.ipynb
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d4e1c4becea54e3d35081c4e2802d466cd0ef6f752ce5a49eeebad3091ef7262
|
| 3 |
+
size 15942
|
requirements.txt
CHANGED
|
@@ -8,6 +8,10 @@ torchmetrics>=0.7.0
|
|
| 8 |
scikit-image
|
| 9 |
pandas
|
| 10 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
# --------- hydra --------- #
|
| 12 |
hydra-core>=1.1.0
|
| 13 |
hydra-colorlog>=1.1.0
|
|
|
|
| 8 |
scikit-image
|
| 9 |
pandas
|
| 10 |
|
| 11 |
+
# --------- libraries for image filters ---------#
|
| 12 |
+
kornia
|
| 13 |
+
|
| 14 |
+
|
| 15 |
# --------- hydra --------- #
|
| 16 |
hydra-core>=1.1.0
|
| 17 |
hydra-colorlog>=1.1.0
|
src/models/focus_traditional.py
CHANGED
|
@@ -7,6 +7,7 @@ from pytorch_lightning import LightningModule
|
|
| 7 |
from torchmetrics import MaxMetric, MeanAbsoluteError, MinMetric
|
| 8 |
from torchmetrics.classification.accuracy import Accuracy
|
| 9 |
import torchvision.models as models
|
|
|
|
| 10 |
|
| 11 |
|
| 12 |
def vol4(img):
|
|
@@ -16,6 +17,31 @@ def vol4(img):
|
|
| 16 |
)
|
| 17 |
|
| 18 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
class TraditionalLitModule(LightningModule):
|
| 20 |
def __init__(
|
| 21 |
self,
|
|
@@ -26,6 +52,7 @@ class TraditionalLitModule(LightningModule):
|
|
| 26 |
|
| 27 |
Args:
|
| 28 |
method (str, optional): The method to use for predicting focus. Defaults to "vol4".
|
|
|
|
| 29 |
|
| 30 |
Raises:
|
| 31 |
Exception: raises exception if method parameter is not known
|
|
@@ -34,6 +61,10 @@ class TraditionalLitModule(LightningModule):
|
|
| 34 |
|
| 35 |
if method == "vol4":
|
| 36 |
self.function = vol4
|
|
|
|
|
|
|
|
|
|
|
|
|
| 37 |
|
| 38 |
def forward(self, x):
|
| 39 |
return self.function(x)
|
|
|
|
| 7 |
from torchmetrics import MaxMetric, MeanAbsoluteError, MinMetric
|
| 8 |
from torchmetrics.classification.accuracy import Accuracy
|
| 9 |
import torchvision.models as models
|
| 10 |
+
import kornia
|
| 11 |
|
| 12 |
|
| 13 |
def vol4(img):
|
|
|
|
| 17 |
)
|
| 18 |
|
| 19 |
|
| 20 |
+
def laplacian(img):
|
| 21 |
+
img_grey = torch.mean(img, dim=0).unsqueeze(0)
|
| 22 |
+
filtered = kornia.filters.laplacian(img_grey, 3)
|
| 23 |
+
mean = torch.mean(filtered)
|
| 24 |
+
return 100 / mean # invert mean to fit metric of lower = better
|
| 25 |
+
|
| 26 |
+
|
| 27 |
+
def midfrequency_dct(img):
|
| 28 |
+
kernel = torch.tensor(
|
| 29 |
+
[
|
| 30 |
+
[
|
| 31 |
+
[1, 1, -1, -1],
|
| 32 |
+
[1, 1, -1, -1],
|
| 33 |
+
[-1, -1, 1, 1],
|
| 34 |
+
[-1, -1, 1, 1],
|
| 35 |
+
]
|
| 36 |
+
]
|
| 37 |
+
)
|
| 38 |
+
|
| 39 |
+
img_grey = torch.mean(img, dim=0).unsqueeze(0)
|
| 40 |
+
filtered = kornia.filters.filter2d(img_grey, kernel)
|
| 41 |
+
sum = torch.sum(filtered)
|
| 42 |
+
return 100 / sum
|
| 43 |
+
|
| 44 |
+
|
| 45 |
class TraditionalLitModule(LightningModule):
|
| 46 |
def __init__(
|
| 47 |
self,
|
|
|
|
| 52 |
|
| 53 |
Args:
|
| 54 |
method (str, optional): The method to use for predicting focus. Defaults to "vol4".
|
| 55 |
+
Possible values are: vol4, mean_laplacian, midfrequency_dct
|
| 56 |
|
| 57 |
Raises:
|
| 58 |
Exception: raises exception if method parameter is not known
|
|
|
|
| 61 |
|
| 62 |
if method == "vol4":
|
| 63 |
self.function = vol4
|
| 64 |
+
if method == "mean_laplacian":
|
| 65 |
+
self.function = laplacian
|
| 66 |
+
if method == "midfrequency_dct":
|
| 67 |
+
self.function = midfrequency_dct
|
| 68 |
|
| 69 |
def forward(self, x):
|
| 70 |
return self.function(x)
|