Object_classification+localization

• Idea

  • Classify and localize 7 dog breeds using resnet + regression heads

• Environment

  • Google colaboratory GPU

• Dataset

  • Chose 7 common dog breeds from Stanford Dogs Dataset. Training on 1193 images

• Training

  • Classification part was inspired by ResNet architecture principles and for localization only changed last resnet layers to regression layers to predict x1, y1, x2, y2 boundary boxes corners coordinates. Training notebook named train.ipynb, testing notebook - test.ipynb

  • Classification head training accuracy during 300 epochs on Google colab GPU
    

  • Classification head training loss during 300 epochs on Google colab GPU
    

  • Regression head training accuracy during 15 epochs on Google colab GPU
    

  • Regression head training loss during 15 epochs on Google colab GPU
    

• Results

---

---

---

---

---

---

---

---