Skip to main content

Measurement Benchmarking

Introduction​

sizeez provides precise body measurements essential for various applications such as apparel sizing, customized clothing, fitness tracking, and more. This documentation offers a thorough overview of the precision evaluation conducted for our body scanning API. It showcases results obtained from both internal controlled tests and real-world scenarios, providing insights into the API's accuracy and reliability.

Internal Controlled Tests​

The following tests are carried out with our internal controlled test set, comprising over 2000 test samples. These tests demonstrate the peak performance achievable by our backend deep learning models.

Test 1: Essential Measurements​

This test case focuses on the four most commonly required body measurements for size recommendation use cases:

  • Chest Girth
  • Waist Girth
  • Maximum Hip Girth
  • Inseam.
Image 1

The error distribution graph visualizes the distribution of percentage errors in the predicted measurements.

Key insights:​

  • A significant peak around 0% error indicates precise measurements for a substantial portion of the samples.
  • The (almost) symmetric distribution suggests that errors are evenly spread around the mean.
  • 96% of all predicted measurements have an absolute error rate lower than 1.5%, highlighting the API's consistency and accuracy.

Test 2: Girth Measurements​

This test case focuses on all the 10 girth measurements:

  • Neck Girth
  • Chest Girth
  • Waist Girth
  • Maximum Hip Girth
  • Upper Arm Girth
  • Elbow Girth
  • Wrist Girth
  • Thigh Girth
  • Calf Girth
  • Ankle Girth
Image 1

The error distribution graph visualizes the distribution of percentage errors in the predicted measurements.

Key insights:​

  • A significant peak around 0% error.
  • The (almost) symmetric distribution.
  • 97% of all predicted measurements have an absolute error rate lower than 2%.

Test 3: Length Measurements​

This test case focuses on all the 6 independent length measurements:

  • Across Back Shoulder Width
  • Shoulder Length
  • Torso Height
  • Upper Arm Length
  • Lower Arm Length
  • Inseam
Image 1

The error distribution graph visualizes the distribution of percentage errors in the predicted measurements.

Key insights:​

  • A significant peak around 0% error.
  • The (almost) symmetric distribution.
  • 96% of all predicted measurements have an absolute error rate lower than 1.5%.
  • In general, the length measurements have higher precision than the girth measurements.

Test 4: All Measurements​

Finally, this test case covers all the measurements:

Image 1

The error distribution graph visualizes the distribution of percentage errors in the predicted measurements.

Key insights:​

  • A significant peak around 0% error.
  • The (almost) symmetric distribution.
  • 97% of all predicted measurements have an absolute error rate lower than 2%.

Real-World Tests​

Below, we present a real-world test case demonstrating the subject taking scans while wearing different kind of outfits. It becomes evident that wearing a skin-tight sport outfit when taking the photo achieves the best scan precision, which is essential for meaningful results.

Test 1: Ideal Conditions: Wearing Skin-Tight Sportswear​

In this test, the subject is outfitted in skin-tight sportswear. In both photos, the subject's silhouette is distinctly distinguishable from the background.

The results indicate that the predicted body measurements have an error rate of under 3%, making them more than adequate for size recommendation purposes.

Image 1

Image 2

Image 3

Absolute Percentage Error

Test 2: Casual Dressing Compromises Scan Accuracy​

In this test, the subject is dressed in a casual fit T-shirt and pants. While his silhouette is clearly distinguishable from the background, it fails to accurately depict his true body shape.

The test result indicates a higher error rate for the predicted body measurements. To ensure precise scans, casual outfit should be avoided.

Image 1

Image 2

Image 3

Absolute Percentage Error

Key Takeaways​

Please consider the following key factors when capturing photos to ensure optimal scan precision:

  • Ensure the subject wears skin-tight clothing that truthfully reflects the body shape.
  • Verify that the subject's silhouette is visually distinguishable from the background.