- Advertisement -

- Advertisement -

Data Augmentation Experimentation

0 10

Get real time updates directly on you device, subscribe now.

- Advertisement -

Looking at the results:

Using Random Rotate:

- Advertisement -

aug_tfms = [Random Rotate(57)]

What the augmentation looks like: Visually the images are rotated randomly at an angle of 57.

Accurate Labels: 55

Accuracy: 67%

Inaccurate Labels: 4

Error: 7%

Using Dihedral:

aug_tfms = [Dihedral()]

What the augmentation looks like: rotates images by random multiples of 90 degrees and/or reflection (flips).

Accurate Labels: 51

Accuracy: 62%

Inaccurate Labels: 9

Error: 18%

Using RandomLighting:

aug_tfms = [RandomLighting(b=0.25,c=0.15)]

What the augmentation looks like: uses 2 arguments b = balance and c= contrast to adjust random picture lighting.

Accurate Labels: 57

Accuracy: 70%

Inaccurate Labels: 12

Error: 21%

Using combination of RandomLighting and RandomDihedral:

aug_tfms = [RandomLighting(b=0.25,c=0.15)] + [RandomDihedral()]

What the augmentation looks like: combination of RandomLighting and RandomDihedral

Accurate Labels: 52

Accuracy: 63%

Inaccurate Labels: 13

Error: 25%

Using a combination of RandomDihedral and RandomRotate:

aug_tfms = [RandomDihedral()] + [RandomRotate(27)]

What the augmentation looks like: combination of RandomDihedral and RandomRotate at a 27 degree angle

Accurate Labels: 56

Accuracy: 68%

Inaccurate Labels: 13

Error: 23%

Using RandomZoomRotate:

aug_tfms = [RandomRotateZoom(deg=45, zoom=2, stretch=1)]

What the augmentation looks like: takes in 3 arguments deg = maximum degrees of rotation, zoom = maximum fraction of zoom and stretch = maximum fraction of stretch

Accurate Labels: 62

Accuracy: 76%

Inaccurate Labels: 8

Error: 13%

Using Padding:

aug_tfms = [AddPadding(pad=50, mode=cv2.BORDER_CONSTANT)]

What the augmentation looks like: takes in 2 arguments = pad = size of padding on top, bottom, left and right and mode = type of cv2 padding modes (CONSTANT, REFLECT, WRAP,

REPLICATE).

Accurate Labels: 54

Accuracy: 66%

Inaccurate Labels: 2

Error: 4%

Using Cutout (Salt & Pepper effect):

aug_tfms = [Cutout(n_holes=200, length=7.5, tfm_y=TfmType.NO)]

What the augmentation looks like: takes in 2 arguments n_holes and length which cuts out n_holes number of square holes of size length in an image at random locations. The holes may be overlapping.

Accurate Labels: 41

Accuracy: 50%

Inaccurate Labels: 9

Error: 22%

- Advertisement -

Get real time updates directly on you device, subscribe now.

- Advertisement -

- Advertisement -

Leave A Reply

Your email address will not be published.

x

We use cookies to give you the best online experience. By agreeing you accept the use of cookies in accordance with our cookie policy.

I accept I decline