Mind boggling that this 1hr lecture is so information dense that the same topics were covered in 3hrs at my university. Shoutout to Prof Pritee Khanna for taking 3hrs to explain this, nptel lectures are very dense and fast-forwarded
@ieatplantsforaliving168 Жыл бұрын
Convolution operation is translation-equivariant, not invariant as the output of the conv. operation changes when an object of interest is translated in the input. CNNs are however a bit more robust to translation, they can learn to be translation invariant (due to pooling layers).
@ronireznik16322 жыл бұрын
In the slide of edge filter there are misleading of opposite operation at at horizontal and vertical gradient.
@SaurabhSingh743 жыл бұрын
There is a typo in the 2d Gaussian function -- 2 is missing from the denominator of the exponent.
@m.d.r30862 жыл бұрын
great the explanation, i want the lectures as pdf plz