The project's goal is to develop a desktop application that uses a CNN model to determine the model of an airplane from a picture. I'll use the FGVC-Aircraft dataset (https://www.robots.ox.ac.uk/vgg/data/fgvc-aircraft/) as my dataset, which has 10000 photos separated into 100 classes.
By far, I was unable to get satisfactory prediction outcomes. I attempted to develop the CNN on my own as well as use common architectures such as AlexNet, EfficientNet, and MobileNet.
- Results of the first CNN (v0.1)
- Results of the modified CNN (v0.2)
- Results of the EfficientNetB0 with imagenet weights
The project is presently on hold due to the difficulties in developing a decent Neural Network capable of handling this dataset.
- Python 3.9
- CUDA v11.2
- cuDNN v8.1.0
- TensorFlow (Keras) 2.9.0