Skip to content

Latest commit

 

History

History
20 lines (13 loc) · 1.28 KB

README.md

File metadata and controls

20 lines (13 loc) · 1.28 KB
Raw

Convolution-and-pooling-from-scratch

This is the work done as a practice purposes and a assignment for a deep learning course where i have to make a forward propagation of a CNN model by making a single layer only.

The dataset used in this work is the same that i have used in my other repository name "Neural network from scratch using NumPy"

here we are given with the 36 filters which are 17x17 dimensional and will be used to detect 36 number of changes in an image like; gradient filter along x-axis or y-axis.

The number of neuron used on hidden and output layers are 128,64 and 10. The results we get are somehow like this.
Results on 500 epochs.
Train accuracy = 72.04% and Testing Accuracy = 70.47%
The number of neuron used on hidden and output layers are 128,64 and 10. The results we get are somehow like this.
Results on 400 epochs and 0.2 learning rate.
Train accuracy = 72.36% and Testing Accuracy = 70.5%

I know the accuracy are not that good but this is just to give you some overview about how to make a simple convolution and pooling functions using numpy.

In case of any query feel free to contact me