"Pet Breed Generative Augmentation Dataset"
Abstract
"Pet Breed Generative Augmentation Dataset is a synthetic image dataset generated to support research on minority-class bias correction in fine-grained animal image classification. It is built on top of the Oxford-IIIT Pet Dataset, which contains 37 cat and dog breeds. Eight breeds were artificially underrepresented to simulate realistic class imbalance conditions, with three breeds reduced to 20 training images and five breeds reduced to 50 training images. Synthetic images were generated for each minority breed using two generative approaches: FastGAN trained independently per breed, and Stable Diffusion 1.5 fine-tuned with Low-Rank Adaptation. Each method produced 500 images per minority breed, giving a total of 8,000 synthetic images across both methods. The dataset is intended for benchmarking generative augmentation strategies under low-data conditions and includes images that exhibit mode collapse artifacts, making it useful for studying failure modes of generative models as well as their successes. All images were generated on a consumer-grade NVIDIA GPU with 6 to 8 gigabytes of memory. Code and generation configurations are publicly available at https:\/\/github.com\/SheshNGupta\/BiasCorrectionInImage."
Read the paper
Related studies
- Learning from First-Generation Qualitative Approaches in the IS Discipline: An Evolutionary View and Some Implications for Authors and Evaluators (PART 1/2)
- Commentary on the clinical and preclinical dosage limits of interstitially administered magnetic fluids for therapeutic hyperthermia based on current practice and efficacy models
- Prosthesis: New Technological Opportunities and Innovative Biomedical Devices
- The history and future of neural modeling for cochlear implants
- Ryanodine receptors: waking up from refractoriness
- Novel Transgenic Rabbit Model Sheds Light on the Puzzling Role of Matrix Metalloproteinase-12 in Atherosclerosis
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.