Shoe-Type-Recognizer

An Image classification Model from Data collection, cleaning, model training as well as deployment and API integration
The final app can classify 36 different types of shoes for both males and females.
Click Here to Visit Web aplication

The App can Provide:

1. The Type of shoes or the name of the shoes
2. The Category of a Shoe(Boots, Heels,Dress shoes .etc)
3. The App can Produce the positive sides and best features as Pros.
4. The App can also produce a negative side and possible problems of shoes as Con.
5. The App can also provide the characteristics and key features of each shoe as Distinction.
   
   Here is a sample of the Final application at work.
   

Types of Shoes and Their Categories are:

Athletic Shoes:

1. Basketball shoes
2. Ballet flats
3. Climbing shoes
4. Golf shoes
5. High-tops shoes
6. Ice-skates shoes
7. Running shoes
8. Soccer shoes

Boots:

1. Army boots
2. Chelsea Boot
3. Cowboy boots
4. Hiking boots
5. Knee high boots
6. Uggs
7. Wellington boots

Cassual Shoes:

1. Chuck Taylor
2. Flip flop
3. Old skool
4. Sandal
5. Sneakers
6. Moccasin
7. Mule shoes

Dress Shoes:

1. Brogues
2. Derby shoe
3. Dress shoe
4. Loafer
5. Oxford shoe

Women's Heels:

1. Cone heels
2. Court shoes
3. High heels
4. Kitten heels
5. Laced booties
6. Lita shoe
7. Mary Jane platforms
8. Platform heels
9. Wedges shoe

Dataset Preparation

Data Collection: The data was collected by downloading from DuckDuckGo search using their term name
DataLoader: Used fastai DataBlock API to set up the DataLoader where Resnet34 was used and was fine-tuned multiple times.
Data Augmentation: fastai provides default data augmentation which operates in GPU.
Details can be found in notebooks/Data_prep_for_capstone2.ipynb

Training and Data Cleaning

Training: Fine-tuned a resnet34 model for 5 epochs initially and then with 2 epochs 4 times and got up to 89.6% accuracy.
Data Cleaning: This part took the highest time. Since the collected data were from the browser, there were many noises. Also, there were images that contained Animations, adds as well as there were garbage data that contains irrelevant subjects. The was cleaned and updated by using fastai library ImageClassifierCleaner.the data were cleaned each time after training and finetuning till the last time which was the final iteration of the model.

Model Deployment

The final model was deployed in HuggingFace Spaces using Gradio App. The implementation can be found in deployment folder or here.

API integration with GitHub Pages

The deployed model API is integrated here in GitHub Pages Website. Implementation and other details can be found in docs folder.