DiverGen reduces distribution bias in instance segmentation by diversifying generative data among models, prompts, and categories.

Check more stories related to machine-learning at: https://hackernoon.com/c/machine-learning.

You can also check exclusive content about #diffusion-models, #instance-segmentation, #data-diversity, #long-tail-recognition, #data-scaling, #computer-vision-pipeline, #clip-inter-similarity, #generative-data-augmentation, and more.

This story was written by: @instancing. Learn more about this writer by checking @instancing's about page,

and for more stories, please visit hackernoon.com.

By introducing Generative Data Diversity Enhancement (GDDE) and conducting a thorough examination of data distribution inconsistencies, DiverGen promotes generative data augmentation for example segmentation. DiverGen recognizes that a lack of real data biases model learning and extends the learnable distribution using three complementary diversity axes: generative model diversity (combining Stable Diffusion and DeepFloyd-IF outputs), prompt diversity (using ChatGPT-generated descriptions), and category diversity (adding ImageNet-based categories).