Implementation of AlexNet from Scratch
This is the tensorflow implementation of AlexNet trained on TPU of Google Colab Pro with 35GB RAM/226GB Disk.
I have used CIFAR-10 dataset for the basic implementation of AlexNet.
The purpose of this implentation is to get a deeper Understanding of the Deep Learning/Computer Vision concepts. Those having a new start in this field can start their practice from this basic implementation and if they have enough resources they can use ImageNet for training as well.
Training:
- loss: 0.0432 - accuracy: 0.9868 - val_loss: 0.7358 - val_accuracy: 0.8227
Test:
- loss: 0.8405 - accuracy: 0.8060
I have got 80.06% accuracy on CIFAR-10 image dataset. This is a quite good accuracy on this quite complex low resolution images.
In case of any questions, you can reach me via Linkedin