[[Paper]](https://arxiv.org/abs/2503.11591) [[Project Page]](https://histodiffusion.github.io/docs/projects/pathae/)

## StabilityAI KL-f8-c16 (SD3.5 VAE) autoencoder finetuned on digital histopathology images.

The original autoencoder is taken from [StabilityAI](https://huggingface.co/stabilityai/stable-diffusion-3.5-large).

We fine-tune the decoder on histopathology images using an L1 reconstruction loss, a PatchGAN adversarial loss and a pathology-specific perceptual loss ([UNI](https://huggingface.co/MahmoodLab/UNI)).