Convolutional Layers 3D

Question 1

In a CNN, a convolutional filter applied to a multi-channel input is itself a…

Accepted Answer

A 3D volume spanning the full input depth — A filter convolves over height, width, AND all input channels at once, summing across depth into one output value per spatial location.

Question 2

Convolving one 3D filter over the input produces…

Accepted Answer

A single 2D feature map, one per filter used — Although the filter is 3D, it collapses the channel dimension via summation, so each filter yields one 2D feature map.

Question 3

To produce a feature VOLUME with depth D, a conv layer uses…

Accepted Answer

D separate, distinct filters — The number of filters sets the output depth; stacking each filter's 2D map forms the output feature volume.

Question 4

The DEPTH of a convolutional feature volume corresponds to…

Accepted Answer

The count of distinct feature channels there — Each channel of the volume is a distinct learned feature map; deeper layers usually increase this channel depth.

Question 5

Visualizing convolution in 3D helps show that the filter…

Accepted Answer

Slides spatially across the full depth — The 3D view clarifies that the kernel moves over height/width but always covers all input channels at each position.

Question 6

True 3D convolution (vs the channel-spanning 2D conv) additionally slides the kernel along…

Accepted Answer

A third spatial or temporal axis — 3D convolutions (used for video, CT/MRI volumes) move the kernel across a depth or time dimension too, capturing spatiotemporal patterns.

Question 7

Weight sharing in convolution means the SAME 3D filter is…

Accepted Answer

Reused at every position there — One filter's weights detect its pattern everywhere in the input, sharply reducing parameters versus dense connections.

Question 8

As you go deeper in a CNN, the spatial dimensions typically ___ while the channel depth ___.

Accepted Answer

shrink; grows — Pooling/striding reduce height×width while more filters increase depth — trading spatial resolution for richer feature representation.

Question 9

Each value in a feature map measures…

Accepted Answer

How strongly the pattern is present there — The convolution's dot product yields an activation indicating the match between the local input region and the filter.

Question 10

The number of parameters in a conv filter is…

Accepted Answer

kH times kW times channels, plus bias — A filter's weight count depends on kernel size and input depth, not the image's spatial size — the key to CNN efficiency.

Simulation

Kernel

Edge Detection

Parameters

Convolve a multi-channel input

Test your knowledge

In a CNN, a convolutional filter applied to a multi-channel input is itself a…

Learn these first

Interview Questions

References & Further Reading

Love this visualization?