Spaces:
Sleeping
Sleeping
Commit
·
d14e266
1
Parent(s):
33289a4
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,13 +1,21 @@
|
|
| 1 |
import json
|
|
|
|
|
|
|
|
|
|
| 2 |
from huggingnft.lightweight_gan.train import timestamped_filename
|
| 3 |
from streamlit_option_menu import option_menu
|
| 4 |
|
| 5 |
from huggingface_hub import hf_hub_download, file_download
|
|
|
|
| 6 |
|
| 7 |
from huggingface_hub.hf_api import HfApi
|
| 8 |
import streamlit as st
|
| 9 |
from huggingnft.lightweight_gan.lightweight_gan import Generator, LightweightGAN, evaluate_in_chunks, Trainer
|
| 10 |
from accelerate import Accelerator
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
|
| 12 |
hfapi = HfApi()
|
| 13 |
model_names = [model.modelId[model.modelId.index("/") + 1:] for model in hfapi.list_models(author="huggingnft")]
|
|
@@ -33,7 +41,7 @@ INTERPOLATION_TEXT = "Text about Interpolation"
|
|
| 33 |
COLLECTION2COLLECTION_TEXT = "Text about Collection2Collection"
|
| 34 |
|
| 35 |
STOPWORDS = ["-old"]
|
| 36 |
-
COLLECTION2COLLECTION_KEYS = ["
|
| 37 |
|
| 38 |
|
| 39 |
def load_lightweight_model(model_name):
|
|
@@ -61,6 +69,11 @@ def clean_models(model_names, stopwords):
|
|
| 61 |
cleaned_model_names.append(model_name)
|
| 62 |
return cleaned_model_names
|
| 63 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 64 |
|
| 65 |
model_names = clean_models(model_names, STOPWORDS)
|
| 66 |
|
|
@@ -141,7 +154,7 @@ if choose == "Generate image":
|
|
| 141 |
nrow=nrows,
|
| 142 |
checkpoint=-1,
|
| 143 |
types=generation_type
|
| 144 |
-
)
|
| 145 |
)
|
| 146 |
|
| 147 |
if choose == "Interpolation":
|
|
@@ -184,13 +197,75 @@ if choose == "Interpolation":
|
|
| 184 |
|
| 185 |
if choose == "Collection2Collection":
|
| 186 |
st.title(choose)
|
| 187 |
-
st.markdown(
|
| 188 |
|
| 189 |
model_name = st.selectbox(
|
| 190 |
'Choose model:',
|
| 191 |
set(model_names) - set(clean_models(model_names, COLLECTION2COLLECTION_KEYS))
|
| 192 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 193 |
generate_image_button = st.button("Generate")
|
| 194 |
|
| 195 |
if generate_image_button:
|
| 196 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import json
|
| 2 |
+
|
| 3 |
+
import torch
|
| 4 |
+
|
| 5 |
from huggingnft.lightweight_gan.train import timestamped_filename
|
| 6 |
from streamlit_option_menu import option_menu
|
| 7 |
|
| 8 |
from huggingface_hub import hf_hub_download, file_download
|
| 9 |
+
from PIL import Image
|
| 10 |
|
| 11 |
from huggingface_hub.hf_api import HfApi
|
| 12 |
import streamlit as st
|
| 13 |
from huggingnft.lightweight_gan.lightweight_gan import Generator, LightweightGAN, evaluate_in_chunks, Trainer
|
| 14 |
from accelerate import Accelerator
|
| 15 |
+
from huggan.pytorch.cyclegan.modeling_cyclegan import GeneratorResNet
|
| 16 |
+
from torchvision import transforms as T
|
| 17 |
+
from torchvision.transforms import Compose, Resize, ToTensor, Normalize, RandomCrop, RandomHorizontalFlip
|
| 18 |
+
from torchvision.utils import make_grid
|
| 19 |
|
| 20 |
hfapi = HfApi()
|
| 21 |
model_names = [model.modelId[model.modelId.index("/") + 1:] for model in hfapi.list_models(author="huggingnft")]
|
|
|
|
| 41 |
COLLECTION2COLLECTION_TEXT = "Text about Collection2Collection"
|
| 42 |
|
| 43 |
STOPWORDS = ["-old"]
|
| 44 |
+
COLLECTION2COLLECTION_KEYS = ["__2__"]
|
| 45 |
|
| 46 |
|
| 47 |
def load_lightweight_model(model_name):
|
|
|
|
| 69 |
cleaned_model_names.append(model_name)
|
| 70 |
return cleaned_model_names
|
| 71 |
|
| 72 |
+
def get_concat_h(im1, im2):
|
| 73 |
+
dst = Image.new('RGB', (im1.width + im2.width, im1.height))
|
| 74 |
+
dst.paste(im1, (0, 0))
|
| 75 |
+
dst.paste(im2, (im1.width, 0))
|
| 76 |
+
return dst
|
| 77 |
|
| 78 |
model_names = clean_models(model_names, STOPWORDS)
|
| 79 |
|
|
|
|
| 154 |
nrow=nrows,
|
| 155 |
checkpoint=-1,
|
| 156 |
types=generation_type
|
| 157 |
+
)[0]
|
| 158 |
)
|
| 159 |
|
| 160 |
if choose == "Interpolation":
|
|
|
|
| 197 |
|
| 198 |
if choose == "Collection2Collection":
|
| 199 |
st.title(choose)
|
| 200 |
+
st.markdown(COLLECTION2COLLECTION_TEXT)
|
| 201 |
|
| 202 |
model_name = st.selectbox(
|
| 203 |
'Choose model:',
|
| 204 |
set(model_names) - set(clean_models(model_names, COLLECTION2COLLECTION_KEYS))
|
| 205 |
)
|
| 206 |
+
nrows = st.number_input("Number of images to generate:",
|
| 207 |
+
min_value=1,
|
| 208 |
+
max_value=10,
|
| 209 |
+
step=1,
|
| 210 |
+
value=1,
|
| 211 |
+
)
|
| 212 |
generate_image_button = st.button("Generate")
|
| 213 |
|
| 214 |
if generate_image_button:
|
| 215 |
+
n_channels = 3
|
| 216 |
+
|
| 217 |
+
image_size = 256
|
| 218 |
+
|
| 219 |
+
input_shape = (image_size, image_size)
|
| 220 |
+
|
| 221 |
+
transform = Compose([
|
| 222 |
+
T.ToPILImage(),
|
| 223 |
+
T.Resize(input_shape),
|
| 224 |
+
ToTensor(),
|
| 225 |
+
Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
|
| 226 |
+
])
|
| 227 |
+
|
| 228 |
+
# generator = modeling_dcgan.Generator.from_pretrained("huggingnft/cryptopunks")
|
| 229 |
+
with st.spinner(text=f"Downloading selected model..."):
|
| 230 |
+
translator = GeneratorResNet.from_pretrained(f'huggingnft/{model_name}',
|
| 231 |
+
input_shape=(n_channels, image_size, image_size),
|
| 232 |
+
num_residual_blocks=9)
|
| 233 |
+
|
| 234 |
+
z = torch.randn(nrows, 100, 1, 1)
|
| 235 |
+
|
| 236 |
+
with st.spinner(text=f"Downloading selected model..."):
|
| 237 |
+
model = load_lightweight_model(f"huggingnft/{model_name.split('__2__')[0]}")
|
| 238 |
+
|
| 239 |
+
with st.spinner(text=f"Generating input images..."):
|
| 240 |
+
punks = model.generate_app(
|
| 241 |
+
num=timestamped_filename(),
|
| 242 |
+
nrow=4,
|
| 243 |
+
checkpoint=-1,
|
| 244 |
+
types="default"
|
| 245 |
+
)[1]
|
| 246 |
+
|
| 247 |
+
pipe_transform = T.Resize((256, 256))
|
| 248 |
+
|
| 249 |
+
input = pipe_transform(punks)
|
| 250 |
+
|
| 251 |
+
with st.spinner(text=f"Generating output images..."):
|
| 252 |
+
output = translator(input)
|
| 253 |
+
|
| 254 |
+
out_img = make_grid(output,
|
| 255 |
+
nrow=4, normalize=True)
|
| 256 |
+
|
| 257 |
+
# out_img = make_grid(punks,
|
| 258 |
+
# nrow=8, normalize=True)
|
| 259 |
+
|
| 260 |
+
out_transform = Compose([
|
| 261 |
+
T.ToPILImage()
|
| 262 |
+
])
|
| 263 |
+
|
| 264 |
+
results = []
|
| 265 |
+
|
| 266 |
+
for out_punk, out_ape in zip(input, output):
|
| 267 |
+
results.append(
|
| 268 |
+
get_concat_h(out_transform(make_grid(out_punk, nrow=1, normalize=True)), out_transform(make_grid(out_ape, nrow=1, normalize=True)))
|
| 269 |
+
)
|
| 270 |
+
for result in results:
|
| 271 |
+
st.image(result)
|