Understanding the Metrics

The current version of Lens AI supports Image data , in the next releases other datatypes like audio, timeseries and text data will be supported.

Image Metrics

Computing image metrics like brightness, sharpness, and noise during inference at regular intervals is crucial for several reasons:

Quality Assurance: Ensures the input images maintain a consistent quality, which is essential for accurate model predictions.
Adaptive Processing: Allows for dynamic adjustments in image preprocessing, enhancing the model's performance in varying conditions.
Anomaly Detection: Identifies anomalies or degradations in image quality that could affect the reliability of the inference results.
Performance Monitoring: Tracks the operational environment, helping to diagnose issues related to image capture or transmission that may impact the overall system performance.
Early Bug Detection: Identifies issues with image quality early, allowing for quicker isolation of the root cause.

Sampling Metrics

Improves Model Performance: Identifies and focuses on the most challenging images, enhancing model accuracy by learning from its mistakes.
Efficient Resource Use: Prioritizes labeling and processing resources on images where the model is least confident, maximizing the value of additional data.
Dynamic Adaptation: Helps the model adapt to changing data distributions over time, maintaining its effectiveness and robustness in real-world applications.

Model Metrics

Performance Monitoring: Ensures the model is running efficiently and meets real-time processing requirements by tracking inference times.
Model Health: Identifies any potential issues or degradations in performance over time, allowing for proactive maintenance and optimization.
Bias Detection: Reveals any biases in class predictions, helping to ensure the model is making balanced and fair predictions across all classes.
Resource Management: Optimizes computational resource allocation by understanding the model's workload and prediction patterns.

PreviousAPI Documentation NextSampling Metrics

Last updated 6 months ago

Image Metrics

Computing image metrics like brightness, sharpness, and noise during inference at regular intervals is crucial for several reasons:

Quality Assurance: Ensures the input images maintain a consistent quality, which is essential for accurate model predictions.

Adaptive Processing: Allows for dynamic adjustments in image preprocessing, enhancing the model's performance in varying conditions.

Anomaly Detection: Identifies anomalies or degradations in image quality that could affect the reliability of the inference results.

Performance Monitoring: Tracks the operational environment, helping to diagnose issues related to image capture or transmission that may impact the overall system performance.

Early Bug Detection: Identifies issues with image quality early, allowing for quicker isolation of the root cause.

Sampling Metrics

Improves Model Performance: Identifies and focuses on the most challenging images, enhancing model accuracy by learning from its mistakes.

Efficient Resource Use: Prioritizes labeling and processing resources on images where the model is least confident, maximizing the value of additional data.

Dynamic Adaptation: Helps the model adapt to changing data distributions over time, maintaining its effectiveness and robustness in real-world applications.

Model Metrics

Performance Monitoring: Ensures the model is running efficiently and meets real-time processing requirements by tracking inference times.

Model Health: Identifies any potential issues or degradations in performance over time, allowing for proactive maintenance and optimization.

Bias Detection: Reveals any biases in class predictions, helping to ensure the model is making balanced and fair predictions across all classes.

Resource Management: Optimizes computational resource allocation by understanding the model's workload and prediction patterns.