Marigold: Affordable Adaptation of Diffusion-Based Image Generators for Image Analysis
Paper • 2505.09358 • Published • 27
Depth Estimation
Diffusers
Safetensors
depth estimation
panoramic images
high resolution
image analysis
computer vision
in-the-wild
zero-shot
import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("prs-eth/PaGeR-depth", dtype=torch.bfloat16, device_map="cuda")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]📟 PaGeR(Panoramic Geometry Estimation) Depth Estimation Model Card
This is a model card for the PaGeR-depth model for monocular depth estimation from a single panoramic ERP image.
The model is fine-tuned from the marigold-e2e-ft model on synthetic [PanoInfinigen](insert link) dataset, as
described in our paper:
- [Paper](paper link) titled "Panorama Geometry Estimation using Single-Step Diffusion Models"
You can also check out other depth and normals models in our collection, or test models in our demo.
Model Details
- Developed by: Vukasin Bozic, Isidora Slavkovic, Dominik Narnhofer, Nando Metzger, Denis Rozumny, Konrad Schindler, Nikolai Kalischek.
- Model type: Generative latent diffusion-based one-step scale-invariant monocular panoramic depth estimation from a single ERP image.
- License: CreativeML OpenRAIL License.
- Model Description: This model can be used to generate an estimated depth map of a panoramic input image.
- Resolution: The model is designed to support large resolutions up to 3K.
- Dataset: PanoInfinigen
- Steps and scheduler: This model works in a swift, one-step regime.
- Outputs:
- Scale-invariant depth map: The predicted values represent a depth map, interpolating between the near and far planes of the model's choice.
- Resources for more information: [Project Website](insert link), [Paper](insert link), Code.
- Cite as:
Add citation
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support