Successful CNN Math Symbol Detection Model

[Madacool01]

After writting the code for the CNN Math Symbol Detection model for my Air-Math project ( #6 ) in Google Colab, the results turned out surprisingly good. Take a look:

Using Google Colab's free NVIDIA T4 GPU, I was able to train this model with 100% accuracy in ~15 min. As seen in the picture, the model had successfully been trained on the dataset from Kagglehub, as we can see it achieves a score of 100% on the Validation Set (20% of the Dataset) on the very first Epoch (The number of times the model will be trained on the dataset). The next step would be to integrate this model into my CV script to detect the symbols, and later send them to Sympy or Wolfram Alpha

[Madacool01]

Just a follow-up, after some research, I found out that getting 100% accuracy on the very first Epoch on a Complex Computer Vision task is practically impossible, and my suspicion was confirmed after I noticed the output:
Found 1 classes: ['extracted_images']...
This indicates that PyTorch's ImageFolder algorithm expects the folder you give to have sub-folders, where each sub-folder is a class.
However, since the Kagglehub saver extracted it into a master folder (which I gave to ImageFolder), it looked inside for class sub-folders, and found "extracted_images", and said:

"Oh! There's only one symbol in this universe, and it's called 'extracted_images'. I will guess that every time!"

Since it had only 1 choice, it was 100% accurate!