The code in this notebook is in response to Siraj's Image classifier Challenge which can be found here.
We intend to perform image classification on the Animals on the Web dataset. The dataset contains images of 10 different classes of animals. However we would perform a binary classification task to classify Leopard and Giraffe images using Transfer Learning in Tensorflow.
The images of these 2 classes of animals are stored in the sub-folders Leopard and Giraffe inside the folder Animals_Data.
The accuracy obtained via transfer learning using Google's Inception-v3 model was approx 98%.
Reference :
Animals on the Web
Tamara L. Berg, David A. Forsyth
Computer Vision and Pattern Recognition (CVPR), 2006
I've already included the preprocessed dataset in the Github repository. But if you want to use other animal images for classification, here's how I did it :
Leopard and Giraffe images can be found in the Animals on the Web Dataset. You can download the Giraffe images (size : 38 MB) from here and the Leopard images (size : 49 MB) from here. I used only the positive samples of these images which can be saved using the possamples.html file. Just extract the two tar files in a folder. Then open the possamples.html for giraffe and leopard in the browser, right click anywhere in the browser window and choose Save as. Save the webpage and automatically all the positive samples of the giraffe and leopard images would be stored in the directory you specified.
- Tensorflow
- PIL (for preprocessing the dataset)
Just run and follow the Ipython Notebook which contains the documented code.
You can also test our animal classification model output_graph.pb by running :
python test.py
After running, for the leopard image which isn't originally in the training set:
The output of the script is as follows:
which shows good confidence in classification.