Spaces:
Runtime error
Runtime error
updated app
Browse files
app.py
CHANGED
|
@@ -1,34 +1,145 @@
|
|
| 1 |
-
import
|
|
|
|
| 2 |
from PIL import Image
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
|
| 4 |
import torch
|
| 5 |
from transformers import AutoImageProcessor, Mask2FormerForUniversalSegmentation
|
| 6 |
from diffusers import StableDiffusionInpaintPipeline
|
| 7 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
|
| 9 |
-
|
|
|
|
|
|
|
| 10 |
|
| 11 |
-
|
| 12 |
-
|
| 13 |
|
|
|
|
|
|
|
| 14 |
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
"runwayml/stable-diffusion-inpainting",
|
| 18 |
-
torch_dtype=torch.float16,
|
| 19 |
-
).to(device)
|
| 20 |
|
|
|
|
|
|
|
| 21 |
|
| 22 |
-
st.
|
| 23 |
|
| 24 |
-
|
| 25 |
-
filename = st.file_uploader("upload an image")
|
| 26 |
-
image = Image.open(filename)
|
| 27 |
-
st.image(image)
|
| 28 |
|
| 29 |
-
#
|
| 30 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 31 |
|
| 32 |
-
|
| 33 |
-
x = st.slider('Select a value')
|
| 34 |
-
st.write(x, 'squared is', x * x)
|
|
|
|
| 1 |
+
import numpy as np
|
| 2 |
+
import pandas as pd
|
| 3 |
from PIL import Image
|
| 4 |
+
from collections import defaultdict
|
| 5 |
+
|
| 6 |
+
import streamlit as st
|
| 7 |
+
from streamlit_drawable_canvas import st_canvas
|
| 8 |
|
| 9 |
import torch
|
| 10 |
from transformers import AutoImageProcessor, Mask2FormerForUniversalSegmentation
|
| 11 |
from diffusers import StableDiffusionInpaintPipeline
|
| 12 |
|
| 13 |
+
import matplotlib as mpl
|
| 14 |
+
|
| 15 |
+
from model import segment_image, inpaint
|
| 16 |
+
|
| 17 |
+
|
| 18 |
+
# define utils and helpers
|
| 19 |
+
DEVICE = 'cuda' if torch.cuda.is_available() else 'cpu'
|
| 20 |
+
|
| 21 |
+
def closest_number(n, m=8):
|
| 22 |
+
""" Obtains closest number to n that is divisble by m """
|
| 23 |
+
return int(n/m) * m
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
def get_mask_from_rectangles(image, mask, height, width, drawing_mode='rect'):
|
| 27 |
+
# Create a canvas component
|
| 28 |
+
canvas_result = st_canvas(
|
| 29 |
+
fill_color="rgba(255, 165, 0, 0.3)",
|
| 30 |
+
stroke_width=2,
|
| 31 |
+
stroke_color="#000000",
|
| 32 |
+
background_image=image,
|
| 33 |
+
update_streamlit=True,
|
| 34 |
+
height=height,
|
| 35 |
+
width=width,
|
| 36 |
+
drawing_mode=drawing_mode,
|
| 37 |
+
point_display_radius=5,
|
| 38 |
+
key="canvas",
|
| 39 |
+
)
|
| 40 |
+
|
| 41 |
+
# get selections from mask
|
| 42 |
+
if canvas_result.json_data is not None:
|
| 43 |
+
objects = pd.json_normalize(canvas_result.json_data["objects"])
|
| 44 |
+
for col in objects.select_dtypes(include=["object"]).columns:
|
| 45 |
+
objects[col] = objects[col].astype("str")
|
| 46 |
+
|
| 47 |
+
if len(objects) > 0:
|
| 48 |
+
# st.dataframe(objects)
|
| 49 |
+
|
| 50 |
+
left_coords = objects.left.to_numpy()
|
| 51 |
+
top_coords = objects.top.to_numpy()
|
| 52 |
+
right_coords = left_coords + objects.width.to_numpy()
|
| 53 |
+
bottom_coords = top_coords + objects.height.to_numpy()
|
| 54 |
+
|
| 55 |
+
# add selections to mask
|
| 56 |
+
for (left, top, right, bottom) in zip(left_coords, top_coords, right_coords, bottom_coords):
|
| 57 |
+
cropped = image.crop((left, top, right, bottom))
|
| 58 |
+
st.image(cropped)
|
| 59 |
+
mask[top:bottom, left:right] = 255
|
| 60 |
+
|
| 61 |
+
st.header("Mask Created!")
|
| 62 |
+
st.image(mask)
|
| 63 |
+
|
| 64 |
+
return mask
|
| 65 |
+
|
| 66 |
+
|
| 67 |
+
def get_mask(image, edit_method, height, width):
|
| 68 |
+
mask = np.zeros((height, width), dtype=np.uint8)
|
| 69 |
+
|
| 70 |
+
if edit_method == "AutoSegment Area":
|
| 71 |
+
|
| 72 |
+
# get displayable segmented image
|
| 73 |
+
seg_prediction, segment_labels = segment_image(image)
|
| 74 |
+
seg = seg_prediction['segmentation'].cpu().numpy()
|
| 75 |
+
viridis = mpl.colormaps.get_cmap('viridis').resampled(np.max(seg))
|
| 76 |
+
seg_image = Image.fromarray(np.uint8(viridis(seg)*255))
|
| 77 |
+
|
| 78 |
+
# display image
|
| 79 |
+
st.image(seg_image)
|
| 80 |
+
|
| 81 |
+
# prompt user to select valid labels to edit
|
| 82 |
+
seg_selections = st.multiselect("Choose segments", zip(segment_labels.keys(), segment_labels.values()))
|
| 83 |
+
if seg_selections:
|
| 84 |
+
tgts = []
|
| 85 |
+
for s in seg_selections:
|
| 86 |
+
tgts.append(s[0])
|
| 87 |
+
|
| 88 |
+
mask = Image.fromarray(np.array([(seg == t) for t in tgts]).sum(axis=0).astype(np.uint8)*255)
|
| 89 |
+
st.header("Mask Created!")
|
| 90 |
+
st.image(mask)
|
| 91 |
+
|
| 92 |
+
elif edit_method == "Draw Custom Area":
|
| 93 |
+
mask = get_mask_from_rectangles(image, mask, height, width)
|
| 94 |
+
|
| 95 |
+
|
| 96 |
+
return mask
|
| 97 |
+
|
| 98 |
+
|
| 99 |
+
|
| 100 |
+
if __name__ == '__main__':
|
| 101 |
+
|
| 102 |
+
st.title("Stable Edit - Edit your photos with Stable Diffusion!")
|
| 103 |
+
|
| 104 |
+
# upload image
|
| 105 |
+
filename = st.file_uploader("upload an image")
|
| 106 |
+
# filename = r"C:\Users\itber\Downloads\Fjord_Cycling.jpg"
|
| 107 |
+
|
| 108 |
+
sf = st.text_input("Please enter resizing scale factor to downsize image (default = 2)", value="2")
|
| 109 |
+
try:
|
| 110 |
+
sf = int(sf)
|
| 111 |
+
except:
|
| 112 |
+
sf.write("Error with input scale factor, setting to default value of 2, please re-enter above to change it")
|
| 113 |
+
sf = 2
|
| 114 |
+
|
| 115 |
+
if filename:
|
| 116 |
+
image = Image.open(filename)
|
| 117 |
|
| 118 |
+
width, height = image.size
|
| 119 |
+
width, height = closest_number(width/sf), closest_number(height/sf)
|
| 120 |
+
image = image.resize((width, height))
|
| 121 |
|
| 122 |
+
st.image(image)
|
| 123 |
+
# st.write(f"{width} {height}")
|
| 124 |
|
| 125 |
+
# Select an editing method
|
| 126 |
+
edit_method = st.selectbox("Select Edit Method", ("AutoSegment Area", "Draw Custom Area"))
|
| 127 |
|
| 128 |
+
if edit_method:
|
| 129 |
+
mask = get_mask(image, edit_method, height, width)
|
|
|
|
|
|
|
|
|
|
| 130 |
|
| 131 |
+
# get inpainted images
|
| 132 |
+
prompt = st.text_input("Please enter prompt for image inpainting", value="")
|
| 133 |
|
| 134 |
+
st.write("Inpainting Images, patience is a virtue :)")
|
| 135 |
|
| 136 |
+
images = inpaint(image, mask, width, height, prompt=prompt, seed=0, guidance_scale=17.5, num_samples=3)
|
|
|
|
|
|
|
|
|
|
| 137 |
|
| 138 |
+
# display all images
|
| 139 |
+
st.write("Original Image")
|
| 140 |
+
st.image(image)
|
| 141 |
+
for i, img in enumerate(images, 1):
|
| 142 |
+
st.write(f"result: {i}")
|
| 143 |
+
st.image(img)
|
| 144 |
|
| 145 |
+
|
|
|
|
|
|
model.py
ADDED
|
@@ -0,0 +1,72 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import torch
|
| 2 |
+
from transformers import AutoImageProcessor, Mask2FormerForUniversalSegmentation
|
| 3 |
+
from diffusers import StableDiffusionInpaintPipeline
|
| 4 |
+
|
| 5 |
+
|
| 6 |
+
device = 'cuda' if torch.cuda.is_available() else 'cpu'
|
| 7 |
+
|
| 8 |
+
|
| 9 |
+
# Image segmentation
|
| 10 |
+
seg_processor = AutoImageProcessor.from_pretrained("facebook/mask2former-swin-base-coco-panoptic")
|
| 11 |
+
seg_model = Mask2FormerForUniversalSegmentation.from_pretrained("facebook/mask2former-swin-base-coco-panoptic")
|
| 12 |
+
|
| 13 |
+
def segment_image(image):
|
| 14 |
+
inputs = seg_processor(image, return_tensors="pt")
|
| 15 |
+
|
| 16 |
+
with torch.no_grad():
|
| 17 |
+
seg_outputs = seg_model(**inputs)
|
| 18 |
+
|
| 19 |
+
# get prediction dict
|
| 20 |
+
seg_prediction = seg_processor.post_process_panoptic_segmentation(seg_outputs, target_sizes=[image.size[::-1]])[0]
|
| 21 |
+
|
| 22 |
+
# get segment labels dict
|
| 23 |
+
segment_labels = {}
|
| 24 |
+
for segment in seg_prediction['segments_info']:
|
| 25 |
+
segment_id = segment['id']
|
| 26 |
+
segment_label_id = segment['label_id']
|
| 27 |
+
segment_label = seg_model.config.id2label[segment_label_id]
|
| 28 |
+
|
| 29 |
+
segment_labels.update({segment_id : segment_label})
|
| 30 |
+
|
| 31 |
+
return seg_prediction, segment_labels
|
| 32 |
+
|
| 33 |
+
# Image inpainting
|
| 34 |
+
# get Stable Diffusion model for image inpainting
|
| 35 |
+
pipe = StableDiffusionInpaintPipeline.from_pretrained(
|
| 36 |
+
"runwayml/stable-diffusion-inpainting",
|
| 37 |
+
torch_dtype=torch.float16,
|
| 38 |
+
).to(device)
|
| 39 |
+
|
| 40 |
+
|
| 41 |
+
def inpaint(image, mask, W, H, prompt="", seed=0, guidance_scale=17.5, num_samples=3):
|
| 42 |
+
""" Uses Stable Diffusion model to inpaint image
|
| 43 |
+
Inputs:
|
| 44 |
+
image - input image (PIL or torch tensor)
|
| 45 |
+
mask - mask for inpainting same size as image (PIL or troch tensor)
|
| 46 |
+
W - size of image
|
| 47 |
+
H - size of mask
|
| 48 |
+
prompt - prompt for inpainting
|
| 49 |
+
seed - random seed
|
| 50 |
+
Outputs:
|
| 51 |
+
images - output images
|
| 52 |
+
"""
|
| 53 |
+
generator = torch.Generator(device="cuda").manual_seed(seed)
|
| 54 |
+
images = pipe(
|
| 55 |
+
prompt=prompt,
|
| 56 |
+
image=image,
|
| 57 |
+
mask_image=mask, # ensure mask is same type as image
|
| 58 |
+
height=H,
|
| 59 |
+
width=W,
|
| 60 |
+
guidance_scale=guidance_scale,
|
| 61 |
+
generator=generator,
|
| 62 |
+
num_images_per_prompt=num_samples,
|
| 63 |
+
).images
|
| 64 |
+
|
| 65 |
+
return images
|
| 66 |
+
|
| 67 |
+
|
| 68 |
+
|
| 69 |
+
|
| 70 |
+
|
| 71 |
+
|
| 72 |
+
|