Tensors represent inputs, model parameters and intermediate values. A scalar has no axes, a vector has one and a matrix has two; image data can add axes for height, width and color channels.
The shape of a tensor describes how its values are arranged, not what those values mean. Software must preserve the intended axis order and interpretation when transforming data between operations.
ELI5
Think of numbers arranged in boxes: a row, a table or a stack of tables. More axes let the arrangement describe more kinds of data.
For example, a color image can store a value for each position and color channel. That is one use of a tensor, not the definition of every tensor.
Is every tensor an image?
No. Tensors also represent text-related values, parameters and many other numerical structures.
What is tensor shape?
Shape specifies the size of each axis in the numerical arrangement.

