Spaces:
Running
Running
Upload 7 files
Browse files- .gitattributes +1 -0
- README.md +1 -1
- app.py +121 -0
- assets/JetBrainsMono-Bold.ttf +0 -0
- assets/ckwb.jpg +0 -0
- assets/dog.jpg +0 -0
- assets/sea.jpg +3 -0
- requirements.txt +7 -0
.gitattributes
CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
36 |
+
assets/sea.jpg filter=lfs diff=lfs merge=lfs -text
|
README.md
CHANGED
@@ -1,6 +1,6 @@
|
|
1 |
---
|
2 |
title: Noisescope
|
3 |
-
emoji:
|
4 |
colorFrom: green
|
5 |
colorTo: yellow
|
6 |
sdk: gradio
|
|
|
1 |
---
|
2 |
title: Noisescope
|
3 |
+
emoji: 😎
|
4 |
colorFrom: green
|
5 |
colorTo: yellow
|
6 |
sdk: gradio
|
app.py
ADDED
@@ -0,0 +1,121 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import uuid
|
2 |
+
from pathlib import Path
|
3 |
+
|
4 |
+
import cv2
|
5 |
+
import gradio as gr
|
6 |
+
import numpy as np
|
7 |
+
import torch
|
8 |
+
from PIL import Image, ImageDraw, ImageFont
|
9 |
+
from diffusers.schedulers import (
|
10 |
+
DDIMScheduler, DDPMScheduler,
|
11 |
+
PNDMScheduler
|
12 |
+
)
|
13 |
+
from gradio.themes import colors
|
14 |
+
from torchvision.transforms.functional import to_pil_image, to_tensor
|
15 |
+
from tqdm import tqdm
|
16 |
+
|
17 |
+
SCHEDULER_CLASSES = {
|
18 |
+
"PNDMScheduler": PNDMScheduler,
|
19 |
+
"DDIMScheduler": DDIMScheduler,
|
20 |
+
"DDPMScheduler": DDPMScheduler,
|
21 |
+
}
|
22 |
+
|
23 |
+
|
24 |
+
@torch.inference_mode()
|
25 |
+
def inference(
|
26 |
+
image: Image.Image,
|
27 |
+
scheduler_name: str,
|
28 |
+
num_timestep: int,
|
29 |
+
fps: int = 5,
|
30 |
+
seed: int = 1117,
|
31 |
+
progress=gr.Progress()
|
32 |
+
):
|
33 |
+
progress(0, "Starting the process...")
|
34 |
+
|
35 |
+
# Set the output directory
|
36 |
+
out_dir = Path('out')
|
37 |
+
if out_dir.exists():
|
38 |
+
# Clean the directory
|
39 |
+
for file in out_dir.iterdir():
|
40 |
+
file.unlink()
|
41 |
+
else:
|
42 |
+
out_dir.mkdir()
|
43 |
+
|
44 |
+
model_id = 'CompVis/stable-diffusion-v1-4'
|
45 |
+
|
46 |
+
scheduler = SCHEDULER_CLASSES[scheduler_name].from_pretrained(model_id, subfolder='scheduler')
|
47 |
+
scheduler.set_timesteps(num_timestep)
|
48 |
+
timesteps = reversed(scheduler.timesteps)
|
49 |
+
|
50 |
+
# Resize image
|
51 |
+
img_width, img_height = image.size
|
52 |
+
image_tensor = to_tensor(image)
|
53 |
+
|
54 |
+
# Font info
|
55 |
+
# Adjust font size based on the image size
|
56 |
+
font_size = int(img_width / 30)
|
57 |
+
font = ImageFont.truetype("assets/JetBrainsMono-Bold.ttf", font_size)
|
58 |
+
|
59 |
+
# Prepare and Add noise
|
60 |
+
generator = torch.Generator().manual_seed(seed)
|
61 |
+
noise = torch.randn(image_tensor.shape, generator=generator)
|
62 |
+
|
63 |
+
# Define the codec and create VideoWriter object
|
64 |
+
fourcc = cv2.VideoWriter_fourcc(*'mp4v') # Be sure to use lower case
|
65 |
+
# fps = 5
|
66 |
+
|
67 |
+
# Create an in-memory buffer
|
68 |
+
out_path = f'{str(out_dir)}/{uuid.uuid4()}.mp4'
|
69 |
+
out = cv2.VideoWriter(out_path, fourcc, fps, (img_width, img_height))
|
70 |
+
|
71 |
+
for i, timestep in enumerate(tqdm(timesteps, desc="Writing video frames")):
|
72 |
+
noised_image = scheduler.add_noise(image_tensor, noise, timestep)
|
73 |
+
noised_image = to_pil_image(noised_image)
|
74 |
+
|
75 |
+
# Add caption to the image
|
76 |
+
draw = ImageDraw.Draw(noised_image)
|
77 |
+
caption = f"Iteration: {i + 1}, Timestep: {timestep.item()}"
|
78 |
+
|
79 |
+
cap_x = img_width / 4
|
80 |
+
cap_y = 10 + (font_size * 2)
|
81 |
+
|
82 |
+
draw.text(xy=(cap_x, cap_y), text=caption, fill=(255, 255, 255), font=font)
|
83 |
+
|
84 |
+
noised_image = cv2.cvtColor(np.array(noised_image), cv2.COLOR_RGB2BGR)
|
85 |
+
out.write(noised_image) # Write out frame to video
|
86 |
+
progress(1 / num_timestep * (i + 1), f"Writing frame {i + 1} of {num_timestep}...")
|
87 |
+
|
88 |
+
out.release() # Release the video writer
|
89 |
+
progress(1, "Video writing completed.")
|
90 |
+
return out_path
|
91 |
+
|
92 |
+
|
93 |
+
if __name__ == '__main__':
|
94 |
+
iface = gr.Interface(
|
95 |
+
fn=inference,
|
96 |
+
inputs=[
|
97 |
+
gr.Image(type="pil", label="Input Image"),
|
98 |
+
gr.Dropdown(
|
99 |
+
choices=[name for name in SCHEDULER_CLASSES.keys()],
|
100 |
+
label="Scheduler",
|
101 |
+
value="PNDMScheduler",
|
102 |
+
),
|
103 |
+
gr.Slider(minimum=10, maximum=990, value=50, step=10, label="Total Timesteps"),
|
104 |
+
gr.Slider(minimum=1, maximum=60, value=5, step=1, label="fps"),
|
105 |
+
gr.Number(value=1117, label="Seed"),
|
106 |
+
],
|
107 |
+
outputs=[
|
108 |
+
gr.PlayableVideo(label="Process of adding noise")
|
109 |
+
],
|
110 |
+
examples=[
|
111 |
+
[Image.open("assets/sea.jpg"), "PNDMScheduler", 999, 1117],
|
112 |
+
[Image.open("assets/ckwb.jpg"), "PNDMScheduler", 30, 1117],
|
113 |
+
[Image.open("assets/dog.jpg"), "PNDMScheduler", 50, 1117],
|
114 |
+
],
|
115 |
+
cache_examples=True,
|
116 |
+
title="NoiseScope 😎",
|
117 |
+
description="Visualize what amount of noise is added at each time step in the forward process of the diffusion model 🧨",
|
118 |
+
theme=gr.Theme(primary_hue=colors.green, secondary_hue=colors.yellow),
|
119 |
+
allow_flagging="never",
|
120 |
+
)
|
121 |
+
iface.queue().launch()
|
assets/JetBrainsMono-Bold.ttf
ADDED
Binary file (278 kB). View file
|
|
assets/ckwb.jpg
ADDED
![]() |
assets/dog.jpg
ADDED
![]() |
assets/sea.jpg
ADDED
![]() |
Git LFS Details
|
requirements.txt
ADDED
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
torch
|
2 |
+
torchvision
|
3 |
+
diffusers
|
4 |
+
transformers
|
5 |
+
safetensors
|
6 |
+
accelerate
|
7 |
+
opencv-python
|