Why processing prepares images for analysis in Computer Vision - Model Pipeline Impact

Model Pipeline - Why processing prepares images for analysis

This pipeline shows how raw images are prepared through processing steps to make them ready for analysis by a machine learning model. Processing cleans and formats images so the model can learn and predict better.

Data Flow - 5 Stages

1Raw Image Input

1000 images x 256 x 256 pixels x 3 color channels→Collect raw images from camera or dataset→1000 images x 256 x 256 pixels x 3 color channels

A photo of a cat with RGB colors

↓

2Resize Images

1000 images x 256 x 256 x 3→Resize all images to 128 x 128 pixels for uniformity→1000 images x 128 x 128 x 3

Cat photo resized smaller to 128x128 pixels

↓

3Normalize Pixel Values

1000 images x 128 x 128 x 3→Scale pixel values from 0-255 to 0-1 range→1000 images x 128 x 128 x 3

Pixel value 128 becomes 0.502

↓

4Data Augmentation

1000 images x 128 x 128 x 3→Apply random flips and rotations to increase data variety→1000 images x 128 x 128 x 3

Cat image flipped horizontally

↓

5Batch Preparation

1000 images x 128 x 128 x 3→Group images into batches of 32 for training→32 images x 128 x 128 x 3 per batch

Batch of 32 cat and dog images ready for model

Training Trace - Epoch by Epoch

Loss
1.2 |****
0.9 |***
0.7 |**
0.5 |*
0.4 |

Epoch	Loss ↓	Accuracy ↑	Observation
1	1.2	0.45	Model starts learning, loss high, accuracy low
2	0.9	0.60	Loss decreases, accuracy improves as model learns features
3	0.7	0.72	Model continues to improve with processed images
4	0.5	0.82	Good convergence, model understands image patterns
5	0.4	0.88	Training stabilizes with high accuracy

Prediction Trace - 4 Layers

Layer 1: Input Image

Layer 2: Resize

Layer 3: Normalize

Layer 4: Model Prediction

Model Quiz - 3 Questions

Why do we resize images before training the model?

A To increase the number of images

B To make all images the same size for consistent input

C To change the image colors

D To remove noise from images

Key Insight

Processing images by resizing and normalizing helps the model learn patterns more easily and improves training performance. Clean, consistent input data is key for good machine learning results.