Sample Redistribution

What is Sample Redistribution?

Sample Redistribution is a technique used in face detection to create more training samples based on the statistics of benchmark datasets. This is done by enlarging the size of square patches cropped from original images during training data augmentation.

How Does Sample Redistribution Work?

During training data augmentation, square patches are cropped from original images using a random size from the set of [0.3,1.0] of the short edge of the original images. To generate more positive samples for stride 8, the random size range is enlarged from [0.3,1.0] to [0.3,2.0]. When the crop box is beyond the original image, average RGB values fill the missing pixels. This technique is used to obtain more training samples for shallow stages.

Why is Sample Redistribution Important?

Efficient face detection under a fixed VGA resolution relies on having enough training samples. In the WIDER FACE dataset, which is commonly used for face detection, most of the faces are smaller than 32x32 pixels. This means that they are predicted by shallow stages, and there is less training data available for these stages. Sample Redistribution helps to address this issue by generating more training samples specifically for shallow stages. This can lead to more accurate and efficient face detection.

Sample Redistribution is an important technique in face detection that can help to improve training data and overall accuracy. By enlarging the size of square patches cropped from original images during training data augmentation, more training samples can be generated specifically for shallow stages. This is important for efficient face detection under fixed VGA resolution and can lead to more accurate and efficient detection overall.

Great! Next, complete checkout for full access to SERP AI.
Welcome back! You've successfully signed in.
You've successfully subscribed to SERP AI.
Success! Your account is fully activated, you now have access to all content.
Success! Your billing info has been updated.
Your billing was not updated.