Passaggio alla visione artificiale: perché i CNN?

Passaggio alla visione artificiale

Oggi passiamo dal trattamento di dati semplici e strutturati utilizzando layer lineari base al fronteggiamento di dati immagine ad alta dimensione. Un'immagine a colori introduce una complessità significativa che le architetture standard non riescono a gestire in modo efficiente. Il deep learning per la visione richiede un approccio specializzato: il Reti neurali convoluzionali (CNN).

1. Perché falliscono le reti completamente connesse (FCN)

In una FCN, ogni pixel di input deve essere collegato a ogni neurone nel livello successivo. Per immagini ad alta risoluzione, ciò causa un'esplodere computazionale, rendendo l'addestramento impossibile e la generalizzazione scarsa a causa di un overfitting estremo.

Input Dimension: A standard $224 \times 224$ RGB image results in $150,528$ input features ($224 \times 224 \times 3$).
Hidden Layer Size: If the first hidden layer uses 1,024 neurons.
Total Parameters (Layer 1): $\approx 154$ million weights ($150,528 \times 1024$) just for the first connection block, requiring massive memory and compute time.

La soluzione CNN

Le CNN risolvono il problema di scalabilità delle FCN sfruttando la struttura spaziale delle immagini. Identificano pattern (come bordi o curve) usando filtri piccoli, riducendo il numero di parametri di diversi ordini di grandezza e promuovendo la robustezza.

TERMINALbash — model-env

> Ready. Click "Run" to execute.

PARAMETER EFFICIENCY INSPECTOR Live

Run comparison to visualize parameter counts.

Question 1

What is the primary benefit of using Local Receptive Fields in CNNs?

Filters only focus on a small, localized region of the input image.

It allows the network to process the entire image globally at once.

It ensures all parameters are initialized to zero.

It removes the need for activation functions.

Question 2

If a $3 \times 3$ filter is applied across an entire image, what core CNN concept is being utilized?

Kernel Normalization

Shared Weights

Full Connectivity

Feature Transposition

Question 3

Which CNN component is responsible for progressively reducing the spatial dimensions (width and height) of the feature maps?

ReLU Activation

Pooling Layers (Subsampling)

Batch Normalization

Challenge: Identifying Key CNN Components

Relate CNN mechanisms to their functional benefits.

We need to build a vision model that is highly parameter efficient and can recognize an object even if it slightly shifts its position in the image.

Step 1

Which mechanism ensures the network can identify a feature (like a diagonal line) regardless of where it is in the frame?

Solution:
Shared Weights. By using the same filter across all locations, the network learns translation invariance.

Step 2

What architectural choice allows a CNN to detect features with fewer parameters than an FCN?

Solution:
Local Receptive Fields (or Sparse Connectivity). Instead of connecting to every pixel, each neuron only connects to a small, localized region of the input.

Step 3

How does the CNN structure lead to hierarchical feature learning (e.g., edges $\to$ corners $\to$ objects)?

Solution:
Stacked Layers. Early layers learn simple features (edges) using convolution. Deeper layers combine the outputs of earlier layers to form complex, abstract features (objects).