dog-vs-cat-mobilenet

dog-vs-cat-classifier-mobilenetv2

This model is a fine-tuned version of pytorch/vision:v0.10.0 MobileNetV2 on a custom Dog vs. Cat image dataset. It achieves the following results on the evaluation set:

• Validation Accuracy: 0.9040
• Training Accuracy: 0.9882
• Final Training Loss: 1.3202

Model description

This model utilizes the MobileNetV2 architecture, a lightweight and efficient convolutional neural network. The classifier "head" was customized with a Dropout layer (p=0.2) and a Linear layer to perform binary classification (Dog vs. Cat). It is designed to run efficiently on CPU environments while maintaining high accuracy.

Intended uses & limitations

Intended Use

• Educational purposes and portfolio demonstration.
• Real-time classification of common household pets (Dogs and Cats).

Limitations

• The model may struggle with low-resolution images or animals in highly crowded backgrounds.
• It is specialized only for two classes; it will not correctly identify other animals.

Training and evaluation data

The model was trained on a balanced dataset of dog and cat images. Data augmentation techniques, including RandomHorizontalFlip and RandomRotation(10), were used to improve generalization and prevent overfitting during the final fine-tuning stage.

Training procedure

Training hyperparameters

The following hyperparameters were used during the final fine-tuning phase:

• optimizer: Adam
• learning_rate: 1e-5 (fine-tuned to 5e-6)
• weight_decay: 0.001
• num_epochs: 3 (final checkpoint)
• batch_size: 32 (standard)
• seed: 42

Training results

Framework versions

• Pytorch 2.10.0+cpu
• Torchvision 0.25.0+cpu