KumaPower's picture
Improve mask editing and add a sample gallery
fd40e27 verified
|
Raw History Blame Contribute Delete
1.43 kB

A newer version of the Gradio SDK is available: 6.32.0

Upgrade
metadata
title: CSA Inpainting Studio
emoji: 🖼️
colorFrom: green
colorTo: blue
sdk: gradio
sdk_version: 6.31.0
python_version: 3.12.12
app_file: app.py
license: cc-by-nc-4.0
short_description: Compare rough and refined image inpainting.
models:
  - KumaPower/CSA
preload_from_hub:
  - >-
    KumaPower/CSA csa-stage2-step97397.pt
    bf82a33eb70a6a0182180ae9f880bb7d1a08c439

CSA Inpainting Studio

Compare two stages of Coherent Semantic Attention for Image Inpainting (ICCV 2019) on the same image and mask. The released bundle pairs rough stage 1 at step 156,181 with CSA refinement at step 97,397.

Upload an image to start painting, or choose an example from the gallery below the workspace. You can also upload a binary mask or sample one with the training recipe. White mask pixels mark holes. Click Compare both stages to view the results side by side or with the wipe slider and download either PNG.

Inputs are resized to 256 × 256 before inference. Both stages preserve known pixels from that resized input. The example gallery uses images from the public CSA and MEDFE releases.

Paper · Source · Weights