Week 7 journal

My (team's) last week goal:

- Using data augmentation techniques, such as rotating or flipping images to help the model to learn more generalizable patterns.

- Creating basic modules to form the same CNN model as my python on cubeIDE with H7 borad.

My work toward goal:

- I defined the types of augmentations I want to apply as arguments to the ImageDataGenerator class. And added the arguments images to original training set.

- Inputed the input tensor, weight tensor, bias tensor, size of the input tensor, and size of the output tensor.

My (team's) next week goal:

-Fix CNN model that used on H7 broad.

Week 7 Milestone:

-By week 7, the convolution layers will be already completed, my algorithm will be able to detect any poker cards.

Compression my progress to the milestone:

-I did not meet my week 7 milestone, my CNN model on H7 broad is not able to detect any poker cards yet. For everything CNN model which trained and tested on Colab is already completed. Compare to my week 7 milestone, the CNN model on Colab is able to detect poker, but H7 broad is printing out wrong predictions, and I compare the feature between Colab and H7 broad of each layers which are different.

Illustrate my Progress:

- Predictions of ace of club poker on Colab:

- Prediction of ace of club o H7, which should be print out that 0 is the max prediction.

- I have learned augmenting the training set to improve the performance of the CNN, the performance of the CNN can be further optimized by tuning the parameters, such as the learning rate, number of layers, filter sizes, define the output image which is the result of the convolution operation.