Publications by authors named "Srikar Yellapragada"
Learned representation-guided diffusion models for large-image generation.
Proc IEEE Comput Soc Conf Comput Vis Pattern Recognit
June 2024
Article Synopsis
- Diffusion models can improve image generation in specialized fields like histopathology and satellite imagery by utilizing self-supervised learning (SSL) embeddings as stand-ins for human labels, which are hard to obtain.
- This new method allows for high-quality images to be created from these embeddings, and it can even generate larger images by combining smaller patches while maintaining their spatial consistency.
- The approach enhances classifier performance on both small patch-level and larger scale classification tasks and shows strong adaptability, successfully working with unseen datasets and different input sources, including text descriptions for image synthesis.
PathLDM: Text conditioned Latent Diffusion Model for Histopathology.
IEEE Winter Conf Appl Comput Vis
January 2024
To achieve high-quality results, diffusion models must be trained on large datasets. This can be notably prohibitive for models in specialized domains, such as computational pathology. Conditioning on labeled data is known to help in data-efficient model training.
View Article and Find Full Text PDF