In this experiment, I will try to master my abilities with Tensorflow on CIFAR-10 dataset.
The CIFAR-10 dataset consists of 60000 32x32 colour images in 10 classes, with 6000 images per class. There are 50000 training images and 10000 test images.
The dataset is divided into five training batches and one test batch, each with 10000 images. The test batch contains exactly 1000 randomly-selected images from each class. The training batches contain the remaining images in random order, but some training batches may contain more images from one class than another. Between them, the training batches contain exactly 5000 images from each class.
The classes are completely mutually exclusive. There is no overlap between automobiles and trucks. "Automobile" includes sedans, SUVs, things of that sort. "Truck" includes only big trucks. Neither includes pickup trucks.
Data could be found at Python CIFAR-10 Data
In this algorithm I had to:
- Import CIFAR-10
- Convert dataset to TFRecord
- Read TFRecord file
- Create batch
- Create a model with 3 Conv Layer and 2 Fully Connected Layers
- Optimizer is Adam with parametric Learning Rate with a decay
- Training for 1000 epochs
- Results written inside Tensorboard for analysis
Open a Terminal and clone this repository.
Go inside the folder.
$ cd <repository_folder>
To run, please insert this code below:
$ python3 main.py --train
To see runtime visualization of how training is going, please type this in Terminal
$ tensorboard --logdir=summary/
In this section, there are illustrated results obtained after a training session.
The comparison between data has been produced using:
$ python3 plot_comparison.py
It needs a CSV file named run_tcomparison-accuracy.csv with 3 columns(step,test,train)
Results are just as expected which means with an accuracy near 40% in training phase compared to the ~10% obtained with random results. In a previous try, I had a 99% which was caused by missing augmentation of data inside the shuffle_batch related to the images and it was suffering of overfitting.