PrakhAI/CatVsDog

CNN Model for binary classification trained on the cats_vs_dogs dataset from tensorflow datasets. Training Colab.

The model was trained for 21h53m on a GCE VM instance (n1-standard-4, 1 x NVIDIA T4) and achieved an accuracy of 74.83% on the test split (10% of the dataset).

The model trained for 90 epochs. Last few epochs' metrics:

After 28122 steps, test loss: 0.5549, test accuracy: 0.7418
After 28449 steps, test loss: 0.5515, test accuracy: 0.7491
After 28776 steps, test loss: 0.5508, test accuracy: 0.7487
After 29103 steps, test loss: 0.5503, test accuracy: 0.7478
After 29430 steps, test loss: 0.5510, test accuracy: 0.7483

Image	Prediction
	Model Prediction - Dog (60.82%), Cat (39.18%)
	Model Prediction - Cat (72.34%), Dog (27.66%)
	Model Prediction - Cat (67.55%), Dog (32.45%)
	Model Prediction - Dog (51.24%), Cat (48.76%)
	Model Prediction - Dog (72.80%), Cat (27.20%)