CNN Explainer

Question 1

CNN Explainer interactively visualizes how a CNN…

Accepted Answer

Classifies an input image by processing it through stacked convolutional layers — It lets you inspect each layer's feature maps and operations on a real image, demystifying how CNNs classify.

Question 2

Hovering over a convolution output in the tool shows that each value comes from…

Accepted Answer

A dot product between a learned filter and a local patch of the previous layer — The visualization animates the filter sliding over local regions, computing the activation at each position.

Question 3

The ReLU layer shown in CNN Explainer…

Accepted Answer

Applies a learnable affine transformation that shifts each activation by a bias — You can see negative values become 0, which makes the network able to learn non-linear functions.

Question 4

The max-pooling step visualized in the tool…

Accepted Answer

Retains only the maximum activation in each spatial window, reducing map size — The animation highlights the largest value per window being kept, reducing spatial size and adding translation invariance.

Question 5

Each FEATURE MAP displayed corresponds to…

Accepted Answer

One filter's activation pattern across the full spatial grid — Brighter regions in a feature map show where that filter detected its pattern in the input.

Question 6

The final output layer in CNN Explainer uses ___ to produce class probabilities.

Accepted Answer

Softmax — Softmax converts the network's final scores into a probability distribution over the image classes.

Question 7

Early convolutional layers in the visualization tend to respond to…

Accepted Answer

Low-level visual patterns such as edges and color gradients — The tool reveals that initial layers detect simple patterns, while deeper layers capture more complex structure.

Question 8

CNN Explainer is based on a small model called…

Accepted Answer

Tiny VGG, a compact two-block convolutional network small enough to run in the browser — It uses a small VGG-style network so all layers and activations can be visualized clearly in the browser.

Question 9

The tool helps reveal that deeper layers have a LARGER…

Accepted Answer

Receptive field, meaning each neuron integrates signals from a wider area of the input — Stacked convolutions and pooling expand the input region each deep neuron depends on, enabling higher-level features.

Question 10

The number of feature maps in a conv layer equals the number of…

Accepted Answer

The number of filters applied in that convolutional layer — Each filter produces one feature map, so the filter count sets the layer's output depth — visible as stacked maps in the tool.

Which layer does what?

Test your knowledge

CNN Explainer interactively visualizes how a CNN…

Learn these first

Interview Questions

References & Further Reading

Love this visualization?