Convolution implementation of sliding windows - what is the tested image

In the introduction to this topic, Professor Ng makes reference to the “tested image” - as shown in yellow under the main image. Could you explain this concept of tested image?

Hi @avnish,

Here Andrew first talks about a “training” dataset having 14x14x3 image dimensions. And later mentions, and explains, how the algo works if your “test” dataset includes 16x16x3 dimension images, rather than similar dimensions of 14x14x3, as included in the “training” dataset.

Thee training dataset is used to first train the model, and concept of test dataset is to test how well the model will perform once it is deployed into ethereal world. The test set might include a different distribution of images as compared with the train set, because we want our model to be robust, and perform well on data it has never seen before.

Hope this explanation helps,