This AI was trained using the power of transfer learning. the base model is mobilenetv3-Large, its classifier head was trained for 15 epochs. it has decent accuracy for its size; Top-1: 79.85% Top-3: 94.53% Top-5: 96.98%. This is achived thanks to heavy data augmentation during training, aswell as lighter augmentation during inference. since its based on Stanford Dogs, it can identify 120 different breeds. it works best in even light with the dogs face or whole body visible.
1 Like