# Image Dataset Drift#

This notebooks provides an overview for using and understanding the image dataset drift check, used to detect drift in simple image properties between train and test datasets.

Structure:

## What Is Image Dataset Drift?#

Drift is simply a change in the distribution of data over time, and it is also one of the top reasons why machine learning model’s performance degrades over time.

Image dataset drift is a drift that occurs in more than one image property at a time, and may even affect the relationships between those properties, which are undetectable by univariate drift methods.

## How Deepchecks Detects Dataset Drift#

This check detects multivariate drift by using a domain classifier. Other methods to detect drift include univariate measures which is used in other checks, such as Image Property Drift check.

## Using Properties to Detect Image Drift#

In computer vision specifically, we can’t measure drift on the images directly, as the individual pixel has little value when estimating drift. Therefore, we calculate drift on different properties of the image, on which we can directly measure drift.

## Which Image Properties Are Used?#

Property name

What is it

Aspect Ratio

Ratio between height and width of image (height / width)

Area

Area of image in pixels (height * width)

Brightness

Average intensity of image pixels. Color channels have different weights according to RGB-to-Grayscale formula

RMS Contrast

Contrast of image, calculated by standard deviation of pixels

Mean Red Relative Intensity

Mean over all pixels of the red channel, scaled to their relative intensity in comparison to the other channels [r / (r + g + b)].

Mean Green Relative Intensity

Mean over all pixels of the green channel, scaled to their relative intensity in comparison to the other channels [g / (r + g + b)].

Mean Blue Relative Intensity

Mean over all pixels of the blue channel, scaled to their relative intensity in comparison to the other channels [b / (r + g + b)].

## Imports#

Note

In this example, we use the pytorch version of the coco dataset and model. In order to run this example using tensorflow, please change the import statements to:

from deepchecks.vision.datasets.detection.coco_tensorflow import load_dataset

import numpy as np

from deepchecks.vision.checks import ImageDatasetDrift


train_ds = load_dataset(train=True, object_type='VisionData')


## Run the check#

### without drift#

check = ImageDatasetDrift()
result = check.run(train_dataset=train_ds, test_dataset=test_ds)
result

Processing Train Batches:
|     | 0/1 [Time: 00:00]
Processing Train Batches:
|#####| 1/1 [Time: 00:00]
Processing Train Batches:
|#####| 1/1 [Time: 00:00]

Processing Test Batches:
|     | 0/1 [Time: 00:00]

Processing Test Batches:
|#####| 1/1 [Time: 00:00]

Processing Test Batches:
|#####| 1/1 [Time: 00:00]

Computing Check:
|     | 0/1 [Time: 00:00]

Computing Check:
|#####| 1/1 [Time: 00:00]

Computing Check:
|#####| 1/1 [Time: 00:00]

Image Dataset Drift

To display the results in an IDE like PyCharm, you can use the following code:

#  result.show_in_window()


The result will be displayed in a new window.

### Insert drift#

Now, we will define a custom collate function that will insert a drift to the training set.

def add_brightness(img):
reverse = 255 - img

drifted_train_ds = load_dataset(train=True, object_type='VisionData')

def created_drifted_collate_function(collate_fn):
def drifted_collate_function(batch):
data_dict = collate_fn(batch)
data_dict['images'] = [add_brightness(np.array(img)) for img in data_dict['images']]
return data_dict
return drifted_collate_function



### Run the check again#

check = ImageDatasetDrift()
result = check.run(train_dataset=drifted_train_ds, test_dataset=test_ds)
result

Processing Train Batches:
|     | 0/1 [Time: 00:00]
Processing Train Batches:
|#####| 1/1 [Time: 00:01]
Processing Train Batches:
|#####| 1/1 [Time: 00:01]

Processing Test Batches:
|     | 0/1 [Time: 00:00]

Processing Test Batches:
|#####| 1/1 [Time: 00:00]

Processing Test Batches:
|#####| 1/1 [Time: 00:00]

Computing Check:
|     | 0/1 [Time: 00:00]

Computing Check:
|#####| 1/1 [Time: 00:00]

Computing Check:
|#####| 1/1 [Time: 00:00]

Image Dataset Drift

## Define a Condition#

Now, we will define a condition that the maximum drift score is less than a certain threshold. In this example we will set the threshold at 0.2. In order to demonstrate the condition, we will use again the original (not drifted) train dataset.

check = ImageDatasetDrift().add_condition_drift_score_less_than(0.2)
result

Processing Train Batches:
|     | 0/1 [Time: 00:00]
Processing Train Batches:
|#####| 1/1 [Time: 00:00]
Processing Train Batches:
|#####| 1/1 [Time: 00:00]

Processing Test Batches:
|     | 0/1 [Time: 00:00]

Processing Test Batches:
|#####| 1/1 [Time: 00:00]

Processing Test Batches:
|#####| 1/1 [Time: 00:00]

Computing Check:
|     | 0/1 [Time: 00:00]

Computing Check:
|#####| 1/1 [Time: 00:00]

Computing Check:
|#####| 1/1 [Time: 00:00]

Image Dataset Drift

Total running time of the script: ( 0 minutes 6.872 seconds)

Gallery generated by Sphinx-Gallery