Fine-tuning a pre-trained ResNet-50 on the Oxford 102 Flowers dataset, using the fastai library. Written for Fellowship.ai's Computer Vision Code Challenge, Cohort 32 application.
Computer Vision (CV) Challenge: Use a pre-trained ResNet50 and train on the Flowers dataset.
- Dataset: Oxford 102 Flower Database
- Source: https://www.robots.ox.ac.uk/~vgg/data/flowers/102
This set contains images of flowers belonging to 102 different categories. Training was done using the fastai library, and the data pre-processing primarily using Pandas.
For this project, I used the images from the file 102flowers.tgz and the images labels imagelabels.mat, available at the link provided. Unzip 102flowers.tgz, and rename the resulting jpg file to unsorted_images. Ensure that this folder, imagelabels.mat, and the Jupyter Notebook are in the same folder before running the code cells.