YOLO concept overview in TensorFlow - Model Metrics & Evaluation

Metrics & Evaluation - YOLO concept overview

Which metric matters for YOLO and WHY

YOLO (You Only Look Once) is a model that finds objects in images. It needs to be fast and accurate at both locating and naming objects. The key metrics are Precision, Recall, and Mean Average Precision (mAP).

Precision tells us how many detected objects are actually correct. This is important to avoid false alarms.

Recall tells us how many real objects the model found. This is important to avoid missing objects.

mAP combines precision and recall across all object classes and detection thresholds. It gives a single score to compare models.

Confusion matrix for object detection

Unlike simple classification, YOLO predicts many boxes. We count True Positives (TP) when a predicted box matches a real object with enough overlap (IoU >= 0.5). False Positives (FP) are wrong boxes, and False Negatives (FN) are missed objects.

    Confusion matrix for one class:

          Predicted
          Object   No Object
    Real
    Object    TP        FN
    No Object FP        TN (not usually counted in detection)
    

Because many boxes are predicted, TN is not usually counted. We focus on TP, FP, and FN to calculate precision and recall.

Precision vs Recall tradeoff in YOLO

If YOLO is set to be very sure before it says "object here," it will have high precision but might miss some objects (low recall).

If YOLO tries to find every object, it will catch more (high recall) but may also make more mistakes (low precision).

Example: In self-driving cars, missing a pedestrian (low recall) is dangerous, so recall is more important. In a photo app tagging objects, precision might be more important to avoid wrong labels.

What good vs bad metrics look like for YOLO

Good metrics:

• Precision and recall both above 0.8 (80%)
• mAP above 0.5 (50%) is considered decent for many tasks
• High recall in safety-critical tasks (e.g., 0.9+)

Bad metrics:

• Precision or recall below 0.5 (50%) means many mistakes or misses
• Very low mAP (below 0.3) means poor detection quality
• High precision but very low recall means many objects are missed

Common pitfalls in YOLO metrics

• Ignoring IoU threshold: Counting detections with low overlap as correct inflates metrics.
• Data leakage: Testing on images seen during training gives unrealistically high scores.
• Overfitting: Very high training mAP but low test mAP means model memorizes training data.
• Imbalanced classes: Rare objects may have low recall but high overall accuracy hides this.
• Not considering speed: YOLO is valued for speed; a slow model with good metrics may not be practical.

Self-check question

Your YOLO model has 98% accuracy but only 12% recall on detecting pedestrians. Is it good for production? Why or why not?

Answer: No, it is not good. The model misses 88% of pedestrians (very low recall), which is dangerous especially in safety-critical uses like self-driving cars. High accuracy here is misleading because most image areas do not contain pedestrians, so the model is mostly correct by saying "no pedestrian" but fails to find actual pedestrians.

Key Result

For YOLO, balancing precision and recall is key; mean Average Precision (mAP) summarizes detection quality across classes.

Practice

1. What is the main idea behind the YOLO (You Only Look Once) model in object detection?

easy

A. It splits the image into a grid and predicts bounding boxes and classes for each cell.

B. It processes each object separately in multiple passes.

C. It uses a sliding window to scan the image multiple times.

D. It detects objects by comparing image patches with a database.

Solution

1. Step 1: Understand YOLO's approach to image processing

   YOLO divides the input image into a grid and predicts bounding boxes and class probabilities for each grid cell in one pass.

2. Step 2: Compare with other detection methods

   Unlike sliding window or multiple passes, YOLO predicts all objects at once, making it fast and efficient.

3. Final Answer:

   It splits the image into a grid and predicts bounding boxes and classes for each cell. -> Option A

4. Quick Check:

   YOLO grid prediction = It splits the image into a grid and predicts bounding boxes and classes for each cell. [OK]

Hint: YOLO predicts all objects in one grid pass [OK]

Common Mistakes:

• Thinking YOLO scans image multiple times
• Confusing YOLO with sliding window methods
• Believing YOLO processes objects one by one

2. Which of the following is the correct way to describe YOLO's output format?

easy

A. A single large tensor containing bounding box coordinates and class probabilities for all grid cells.

B. A list of detected objects with coordinates and classes, one by one.

C. Multiple small tensors each representing a detected object separately.

D. A heatmap showing object presence without bounding boxes.

Solution

1. Step 1: Recall YOLO's output structure

   YOLO outputs a single large tensor that encodes bounding boxes and class probabilities for every grid cell simultaneously.

2. Step 2: Eliminate other options

   Options A and C describe separate outputs per object, which is not how YOLO works. A heatmap showing object presence without bounding boxes describes a heatmap, which is not YOLO's output.

3. Final Answer:

   A single large tensor containing bounding box coordinates and class probabilities for all grid cells. -> Option A

4. Quick Check:

   YOLO outputs one big tensor = A single large tensor containing bounding box coordinates and class probabilities for all grid cells. [OK]

Hint: YOLO outputs one big tensor for all predictions [OK]

Common Mistakes:

• Thinking YOLO outputs separate tensors per object
• Confusing YOLO output with heatmaps
• Assuming YOLO outputs a list instead of a tensor

3. Given a YOLO model that splits an image into a 7x7 grid and predicts 2 bounding boxes per cell with 20 classes, what is the shape of the output tensor?

medium

A. (7, 7, 28)

B. (7, 7, 50)

C. (7, 7, 44)

D. (7, 7, 30)

Solution

1. Step 1: Calculate output features per grid cell

   Each cell predicts 2 bounding boxes. Each box has 5 values (x, y, width, height, confidence), so 2*5=10. Plus 20 class probabilities, total 10+20=30 per cell.

2. Step 2: Calculate total output tensor shape

   Grid size is 7x7, so output shape is (7, 7, 30). But (7, 7, 50) says (7,7,50), so check carefully.

3. Step 3: Re-examine calculations

   Actually, each bounding box has 5 values, so 2 boxes = 10 values. Adding 20 class probabilities gives 30 per cell, so output shape is (7,7,30).

4. Final Answer:

   (7, 7, 30) -> Option D

5. Quick Check:

   Output shape = grid size * (2*5 + 20) = (7,7,30) [OK]

Hint: Output depth = 2*5 + number_of_classes [OK]

Common Mistakes:

• Adding bounding boxes and classes incorrectly
• Confusing number of boxes with classes
• Miscounting bounding box parameters

4. You wrote a YOLO model that outputs a tensor of shape (7, 7, 25) but your dataset has 3 bounding boxes per cell and 20 classes. What is the likely error?

medium

A. Number of classes should be 5, not 20.

B. Output tensor depth is too small; should be 3*5 + 20 = 35, not 25.

C. Output tensor shape is correct; no error.

D. Grid size should be 5x5, not 7x7.

Solution

1. Step 1: Calculate expected output depth

   Each bounding box has 5 values, so 3 boxes = 15 values. Adding 20 classes gives 15 + 20 = 35 values per grid cell.

2. Step 2: Compare with given output shape

   The model outputs (7, 7, 25), which is less than expected 35, indicating the output tensor depth is too small.

3. Final Answer:

   Output tensor depth is too small; should be 3*5 + 20 = 35, not 25. -> Option B

4. Quick Check:

   Output depth mismatch = Output tensor depth is too small; should be 3*5 + 20 = 35, not 25. [OK]

Hint: Check output depth = boxes*5 + classes [OK]

Common Mistakes:

• Ignoring bounding box parameters count
• Assuming output shape is always (7,7,25)
• Mixing grid size with output depth

5. You want to improve YOLO's real-time detection speed on a mobile device. Which approach is best while keeping reasonable accuracy?

hard

A. Reduce the grid size to 3x3 and keep 2 bounding boxes per cell.

B. Increase the number of bounding boxes per cell to 5 for better coverage.

C. Use a smaller backbone network and reduce input image size.

D. Increase input image size to get more details.

Solution

1. Step 1: Understand speed vs accuracy trade-offs

   Reducing model size and input resolution speeds up inference but may reduce accuracy slightly.

2. Step 2: Evaluate options for mobile optimization

   Reduce the grid size to 3x3 and keep 2 bounding boxes per cell reduces grid size too much, hurting detection detail. Increase the number of bounding boxes per cell to 5 for better coverage increases complexity, slowing down. Increase input image size to get more details increases input size, slowing down.

3. Step 3: Choose best option

   Using a smaller backbone and smaller input size balances speed and accuracy well for mobile devices.

4. Final Answer:

   Use a smaller backbone network and reduce input image size. -> Option C

5. Quick Check:

   Smaller model + input = faster detection [OK]

Hint: Smaller model and input size boost speed [OK]

Common Mistakes:

• Reducing grid size too much loses detail
• Increasing bounding boxes slows model
• Increasing input size hurts speed