ariel415el / gpdm

Generating Natural Images with Direct Patch Distribution Matching

  • Public
  • 5.8K runs
  • T4
  • GitHub
  • Paper
  • License
Iterate in playground

Input

string

(Reshuffle): Start from a noisy and generates samples from the same scene in the reference image. (Retarget): Start from a stretched version of the reference and recreates the scene in another aspect ratio. (Style-transfer): Recreates a content-image with the style of the reference image

Default: "Reshuffle"

integer

How many output images to generate. Generating multiple images at once improves the quality and diversity of the results

Default: 4

integer

Number of ramdom projections for SWD. More is better results but is slower and memory inefficient

Default: 64

integer

Size of the extracted a patches

Default: 8

integer

Change the output's aspect ratio (Retargeting)

Default: 1

integer

Change the output's aspect ratio (Retargeting)

Default: 1

*file
Preview
reference_image

The main input image - Style image in style-transfer.

file
Preview
content_image

Only relevant for style-transfer

Output

output
Generated in

This output was created using a different version of the model, ariel415el/gpdm:e07faab3.

Run time and cost

This model costs approximately $0.053 to run on Replicate, or 18 runs per $1, but this varies depending on your inputs. It is also open source and you can run it on your own computer with Docker.

This model runs on Nvidia T4 GPU hardware. Predictions typically complete within 4 minutes. The predict time for this model varies significantly based on the inputs.

Readme

GPDM live demo

A live demo for GPDM, an algorithm introduced in “Generating Natural Images with Direct Patch Distribution Matching”

This demo allows performing the “Reshuffle”, “Retarget” or “Style-transfer” tasks on any uploaded image online with a considerable amount of control over the GPDM algorithm hyper-parameters

Cite

@article{elnekave2022generating, title={Generating natural images with direct Patch Distributions Matching}, author={Elnekave, Ariel and Weiss, Yair}, journal={arXiv preprint arXiv:2203.11862}, year={2022} }