This is a Classification project. A Web Application that receives a dog picture as input and returns its breed as output. The model predicts the breed from 150 different breeds. CLICK HERE TO START THE APP
LIVE DEMO   Computer Preview  Â
Mobile Preview  Â
Setup  Â
Web Application Section  Â
AI Section
  Â
First install python3.7 and python3.7-venv on the OS.
By typing python3.7 and hiting Enter, python shell will appear (version 3.7).
Leave the python shell and run the commands below:
$ git clone https://github.com/amirdy/dog-breed-classification.git
$ cd dog-breed-classification/Web_App
$ python3.7 -m venv env
$ source ./env/bin/activate
(env)$ pip install -r requirements.txt
(env)$ pip install torch==1.6.0+cpu torchvision==0.7.0+cpu -f https://download.pytorch.org/whl/torch_stable.html
(env)$ pip install efficientnet_pytorch==0.6.3
(env)$ flask run
It will run a Flask development server.
For enabling Fontawesome icons, you must add Fontawesome CDN into the index.html ( inside the <head> tag ).
You can find this CDN from here.
HTML - CSS - Js - Jquery - Bootstrap - Chart.js - Flask - Python
Combination of 3 datasets:
Number of total images:
- 22790
-
Using pretrained EfficientNet-B31 model2
-
Image was fed into the model and the last Conv features were saved
-
These features were fed into a one layer classifier (Training)
-
- Drop out (0.8) [in classifier] + 5 Fold Cross-Validation
-
- 0.8
-
- 25
-
- 0.0001
-
- ADAM
-
- Approximately : 0.8 | 0.2
-
- Approximately : 0.8 | 0.2
-
- Python - Pytorch ( Using Google Colab Pro )
-
- Resize (H = 400, W = 350) | Rotate | Normalize
-
- loss : 0.2281 | Accuracy: 92.5884 See Details
-
-
- Best Validation Loss : 0.2477 - (In Epoch 134 | Accuracy : 92.60 %) [This model is selected] - Test Accuracy on this model : 92.48 %
-
-
-
- Best Validation Loss : 0.2262 - (In Epoch 147 | Accuracy : 92.39 %) [This model is selected] - Test Accuracy on this model : 92.68 %
-
-
-
- Best Validation Loss : 0.2489 - (In Epoch 111 | Accuracy : 91.86 %) [This model is selected] - Test Accuracy on this model : 92.39 %
-
-
-
- Best Validation Loss : 0.2630 - (In Epoch 134 | Accuracy : 91.59 %) [This model is selected] - Test Accuracy on this model : 92.46 %
-
-
-
- Best Validation Loss : 0.2425 - (In Epoch 135 | Accuracy : 92.46 %) [This model is selected] - Test Accuracy on this model : 92.33 %
-
