Annotation quality in Computer Vision - Model Pipeline Trace

Model Pipeline - Annotation quality

This pipeline shows how annotation quality affects a computer vision model. It starts with images and their labels, checks annotation accuracy, trains a model, and measures how well it learns and predicts.

Data Flow - 5 Stages

1Raw images with annotations

1000 images x 256x256 pixels→Collect images and label bounding boxes or masks→1000 images x 256x256 pixels with annotations

Image of a cat with a bounding box drawn around it

↓

2Annotation quality check

1000 images with annotations→Review and correct annotation errors (missing or wrong labels)→1000 images with cleaned annotations

Corrected bounding box around the cat, no missing labels

↓

3Data preprocessing

1000 images with cleaned annotations→Resize images, normalize pixels, encode labels→1000 images x 224x224 pixels, normalized, with encoded labels

Image resized to 224x224, pixel values scaled 0-1

↓

4Model training

800 training images x 224x224 pixels with labels→Train convolutional neural network to detect objects→Trained model

Model learns to detect cats and dogs from images

↓

5Model evaluation

200 test images x 224x224 pixels with labels→Calculate accuracy and loss on test set→Accuracy and loss metrics

Model achieves 85% accuracy on test images

Training Trace - Epoch by Epoch

Loss
1.2 |*       
0.9 | *      
0.7 |  *     
0.5 |   *    
0.4 |    *   
    +---------
     1 2 3 4 5 Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	1.2	0.45	Model starts learning but accuracy is low due to noisy annotations
2	0.9	0.6	Loss decreases and accuracy improves as model learns patterns
3	0.7	0.72	Better annotation quality helps model improve faster
4	0.5	0.8	Model converges with good accuracy on clean data
5	0.4	0.85	Final epoch shows stable loss and high accuracy

Prediction Trace - 5 Layers

Layer 1: Input image

Layer 2: Convolutional layers

Layer 3: Fully connected layers

Layer 4: Softmax activation

Layer 5: Prediction output

Model Quiz - 3 Questions

What happens to model accuracy when annotation quality improves?

A Accuracy decreases

B Accuracy increases steadily

C Accuracy stays the same

D Accuracy becomes random

Key Insight

Good annotation quality is crucial for training effective computer vision models. Clean, accurate labels help the model learn faster and achieve higher accuracy, while noisy or wrong annotations slow learning and reduce performance.