Spaces:
Sleeping
Sleeping
File size: 2,833 Bytes
e2e2760 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 |
import gradio as gr
import cv2
import os
def restore(image_missing, image_inference, channel):
if len(image_missing.shape) != 3 or len(image_inference.shape) != 3:
return None #[None, "Error: Please provide RGB images!"]
c = { "R": 0, "G": 1, "B": 2 }[channel] # cv2 uses BGR order
image_missing[:,:,c] = image_inference[:,:,c]
return image_missing
examples=[
[os.path.join(os.path.dirname(__file__), "missing/bird_r.webp" ), os.path.join(os.path.dirname(__file__), "inference/bird_r.webp" ), "R"],
[os.path.join(os.path.dirname(__file__), "missing/bird_g.webp" ), os.path.join(os.path.dirname(__file__), "inference/bird_g.webp" ), "G"],
[os.path.join(os.path.dirname(__file__), "missing/bird_b.webp" ), os.path.join(os.path.dirname(__file__), "inference/bird_b.webp" ), "B"],
[os.path.join(os.path.dirname(__file__), "missing/dog2_r.webp" ), os.path.join(os.path.dirname(__file__), "inference/dog2_r.webp" ), "R"],
[os.path.join(os.path.dirname(__file__), "missing/dog2_g.webp" ), os.path.join(os.path.dirname(__file__), "inference/dog2_g.webp" ), "G"],
[os.path.join(os.path.dirname(__file__), "missing/dog2_b.webp" ), os.path.join(os.path.dirname(__file__), "inference/dog2_b.webp" ), "B"],
[os.path.join(os.path.dirname(__file__), "missing/house2_r.webp"), os.path.join(os.path.dirname(__file__), "inference/house2_r.webp"), "R"],
[os.path.join(os.path.dirname(__file__), "missing/house2_g.webp"), os.path.join(os.path.dirname(__file__), "inference/house2_g.webp"), "G"],
[os.path.join(os.path.dirname(__file__), "missing/house2_b.webp"), os.path.join(os.path.dirname(__file__), "inference/house2_b.webp"), "B"],
]
app = gr.Interface(
fn=restore,
inputs=[
gr.Image("missing/dog2_r.webp", type="numpy", label="Missing channel"),
gr.Image("inference/dog2_r.webp", type="numpy", label="Inference"),
gr.Radio(["R", "G", "B"], label="Restore channel"),
],
outputs="image",
examples=examples,
title="Restore missing RGB channel",
description="Restore missing channel of a RGB image by using a ControlNet to guide image generation of Stable Diffusion to infer missing channel from the other two channels https://huggingface.co/GeroldMeisinger/control-channels . Examples are missing one channel each (use `decompose.py`) and inference examples are cherry-picked from generation with ControlNet. The submit button replaces the missing channel from the inference. To generate your own inferences you have to set up Stable Diffusion with ControlNet 'channels-rgb'. Note: While the inference images look fine on their own and you might as well use them, we are only interested in one of their channels.")
if __name__ == "__main__":
app.launch()
|