Footwear Image-classification-using-ConvNet And Visually-similar-Recommendation

Dataset

We created our own dataset which contain 30,000 footwear images sourced from
Flipkart, Snapdeal and Amazon. We divided our datasets into twelve class labels.

• sandals-floaters
• slippers­flipflop
• sneakers
• sports
• formal
• loafers
• ethnic
• mens­boot
• flats
• heels
• ballerinas
• womens­boot

Download Dataset

Reference

http://cs231n.stanford.edu/reports/nealk_final_report.pdf

Steps to run the Project

1. Create a virtualenv and activate it

virtualenv myvenv

source myvenv/bin/activate 2. pip install -r requirement.txt 3. Run topick.py to convert the dataset in pickel format 4. open ipython notebook 5. create a new notebook and run following commands (shift + enter to run notebook's cell code) 6. from input_output import Image_classification 7. obj = Image_classification() 8. obj.training() 9. obj.extract_features()

output of last fully-connected layer 10. obj.testing()
11. obj.visually_similar()

Demo