MNIST, FMNIST, KMNIST, Music Genre DataSet for CVCNN Paper

The datasets used in this work include MNIST, Fashion-MNIST, KMNIST, and two music genre datasets curated for audio classification tasks. MNIST is a benchmark dataset of 70,000 grayscale handwritten digits (0\u20139), each of size 28\u00d728 pixels, widely used in evaluating image classification models [1]. Fashion-MNIST is a modern alternative, containing grayscale images of clothing items across 10 categories [2], while KMNIST provides images of Japanese Kuzushiji characters, offering a more complex recognition task beyond Latin alphabets [3]. In addition, we include two audio-based datasets: a Music Genre Binary Dataset with samples from two genres (classical and rock), and a Music Genre Multiclass Dataset spanning several music genres [4]. Both music datasets are stored as WAV files and can be preprocessed into spectrogram or MFCC features for machine learning. Collectively, these datasets support reproducibility in evaluating deep learning models across vision and audio domains, enabling consistent benchmarking for classification tasks.References:[1] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, \u201cGradient-based learning applied to document recognition,\u201d Proceedings of the IEEE, vol. 86, no. 11, pp. 2278\u20132324, 1998.[2] H. Xiao, K. Rasul, and R. Vollgraf, \u201cFashion-mnist: a novel image dataset for benchmarking machine learning algorithms,\u201d 2017.[3] T. Clanuwat, M. Bober-Irizar, A. Kitamoto, A. Lamb, K. Yamamoto, and D. Ha, \u201cDeep learning for classical japanese literature,\u201d 2018.[4] G. Tzanetakis and P. Cook, \u201cMusical genre classification of audio signals,\u201d IEEE Transactions on Speech and Audio Processing, 10(5):293\u2013302, 2002.