CV project workflow in Computer Vision - Model Metrics & Evaluation

Metrics & Evaluation - CV project workflow

Which metric matters for CV project workflow and WHY

In computer vision projects, the choice of metric depends on the task. For image classification, accuracy is common because it shows how many images were correctly labeled. For object detection, mean Average Precision (mAP) is key as it measures how well the model finds and labels objects. For segmentation, Intersection over Union (IoU) tells how well predicted areas match the true areas. Choosing the right metric helps you know if your model is truly learning what matters.

Confusion matrix example for image classification

      | Predicted Cat | Predicted Dog |
      |--------------|---------------|
      | True Cat: 50 | False Dog: 5  |
      | False Cat: 3 | True Dog: 42  |

      Total samples = 50 + 5 + 3 + 42 = 100

      Precision (Cat) = TP / (TP + FP) = 50 / (50 + 5) = 0.909
      Recall (Cat) = TP / (TP + FN) = 50 / (50 + 3) = 0.943
    

This matrix helps you see where the model makes mistakes and calculate metrics.

Precision vs Recall tradeoff with examples

Imagine a face recognition system for phone unlock:

• High precision: The system rarely lets strangers in (few false accepts), but might sometimes not recognize the owner (false rejects).
• High recall: The system always recognizes the owner (few false rejects), but might sometimes let strangers in (false accepts).

Depending on what matters more (security or convenience), you adjust the model to favor precision or recall.

What good vs bad metric values look like in CV projects

For image classification:

• Good: Accuracy above 90%, precision and recall balanced above 85%
• Bad: Accuracy below 70%, or very low recall meaning many true objects are missed

For object detection:

• Good: mAP above 0.7 means the model finds and labels objects well
• Bad: mAP below 0.4 means poor detection and many missed objects

Common pitfalls in CV metrics

• Accuracy paradox: High accuracy can be misleading if classes are imbalanced (e.g., many background images, few objects).
• Data leakage: Using test images in training inflates metrics falsely.
• Overfitting: Very high training accuracy but low test accuracy means the model memorizes instead of learning.
• Ignoring metric choice: Using accuracy for detection tasks can hide poor localization performance.

Self-check question

Your image classifier has 98% accuracy but only 12% recall on a rare class. Is it good for production?

Answer: No. The model misses most examples of the rare class (low recall), which can be critical depending on the task. High accuracy is misleading if the rare class is important.

Key Result

Choosing the right metric like accuracy, mAP, or IoU is key to correctly evaluate computer vision models and avoid misleading results.