Convolutional classification model with an accuracy of ~90%.
CIFAR is an acronym that stands for the Canadian Institute For Advanced Research. The dataset is comprised of 60,000 32×32 pixel color photographs of objects from 10 classes. The class labels and their standard associated integer values are listed below.
0: airplane
1: automobile
2: bird
3: cat
4: deer
5: dog
6: frog
7: horse
8: ship
9: truck
- Tkinter for GUI.
- Keras for Deep Learning.
As shown below, these are screenshots of models that gave unsatisfactory results upon training. Basically, there were two cases that were not favourable :
1. Accuracy was stuck at ~75%.
2. After a set of epochs overfitting occured.
Case 1 : As you can see below, after around 40 epochs, overfitting occured :
Case 2 : Learnt from my mistakes above, I tried the following, but in this case the accuracy was stuck at ~80%.
When I incresed number of epochs, the accuracy further decresed.
Case 3 : Again, as shown below, when I added Kernel Regularization to overcome above scenario, overfitting occured.
There are few more cases that I tried training upon. But all were failed due to similar reasons.
Model 1 : I used Relu activation function and ADAM optimizer with 125 number of epochs
with decresing Learning rate and voila! I got a stable accuracy of ~90%.
If you take a closer look at figure above,
you will observe an increment in accuracy at epoch 75 as well as at 100.
This is were I decreased my learning rate.
This is the model that is used in project.
I tried to increase the accuracy by incresing number of epochs to 175
and further decresing Learning rate, but results were not as I expected them to be.
You can observe after 125 epoch overfitting started to occur.
Model 2 : Here, I used Elu activation function and RMSProp optimizer
with same 125 number of epochs with decresing Learning rate.