Benchmark analysis of various pre-trained deep learning models on ASSIRA cats and dogs dataset

Abstract Image classification using deep learning has gained significant attention, with various datasets available for benchmarking algorithms and pre-trained models. This study focuses on the Microsoft ASIRRA dataset, renowned for its quality and benchmark standards, to compare different pre-trained models. Through experimentation with optimizers, loss functions, and hyperparameters, this research aimed to enhance model performance. Notably, this study achieved significant accuracy improvements with minimal modifications to the training process. Experiments were conducted across three computer architectures, yielding superior accuracy results compared to previous studies on this dataset. The NASNet Large model emerged with the highest accuracy at 99.65%. The findings of this research demonstrate the effectiveness of hyperparameter tuning for renowned pre-trained models, suggesting optimal settings for improved classification accuracy. This study underscores the potential of deep learning approaches in achieving superior performance by hyperparameter tuning for image classification tasks.