interface files
Browse files- scripts/app.py +52 -0
- scripts/generate.py +167 -0
scripts/app.py
ADDED
@@ -0,0 +1,52 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# app.py
|
2 |
+
import streamlit as st
|
3 |
+
from generate import generate_image
|
4 |
+
import os
|
5 |
+
|
6 |
+
# Set up page config
|
7 |
+
st.set_page_config(page_title="Visual Reconstruction from Brain", layout="centered")
|
8 |
+
|
9 |
+
# Configure Streamlit to use the correct host
|
10 |
+
import streamlit.web.cli as stcli
|
11 |
+
import sys
|
12 |
+
sys.argv = ["streamlit", "run", "scripts/app.py", "--server.address", "10.192.12.247", "--server.port", "8501", "--browser.serverAddress", "10.192.12.247"]
|
13 |
+
|
14 |
+
st.title("🧠 Imagine an Image!")
|
15 |
+
|
16 |
+
# Subject selection
|
17 |
+
sub = st.selectbox("Select Subject", options=[1, 2, 5, 7], index=0)
|
18 |
+
|
19 |
+
# Image ID input
|
20 |
+
image_id = st.number_input("Enter Image ID", min_value=0, step=1)
|
21 |
+
|
22 |
+
original_path = f'data/nsddata_stimuli/test_images/{image_id}.png'
|
23 |
+
if os.path.exists(original_path):
|
24 |
+
st.image(original_path, caption="Original Image", use_column_width=True)
|
25 |
+
else:
|
26 |
+
st.warning("Original image not found.")
|
27 |
+
# Text prompt
|
28 |
+
annot = st.text_input("Describe what you imagined", placeholder="e.g., a dog under a tree")
|
29 |
+
|
30 |
+
# Parameters
|
31 |
+
strength = st.slider("Diffusion Strength", 0.0, 1.0, 0.75, 0.05)
|
32 |
+
mixing = st.slider("Mixing Strength", 0.0, 1.0, 0.4, 0.05)
|
33 |
+
|
34 |
+
# Submit button
|
35 |
+
if st.button("Reconstruct Image"):
|
36 |
+
with st.spinner("Reconstructing... please wait"):
|
37 |
+
try:
|
38 |
+
original_path, imagined_path = generate_image(sub, image_id, annot, strength, mixing)
|
39 |
+
# if os.path.exists(original_path):
|
40 |
+
# st.image(original_path, caption="Original Image", use_column_width=True)
|
41 |
+
# else:
|
42 |
+
# st.warning("Original image not found.")
|
43 |
+
if os.path.exists(imagined_path):
|
44 |
+
st.image(imagined_path, caption="Imagined Reconstruction", use_column_width=True)
|
45 |
+
else:
|
46 |
+
st.warning("Imagined image not found.")
|
47 |
+
except Exception as e:
|
48 |
+
st.error(f"⚠️ Error during generation: {e}")
|
49 |
+
|
50 |
+
# Optional: For cloud users
|
51 |
+
st.markdown("---")
|
52 |
+
# st.markdown("🔗 Access the app at: http://10.192.12.247:8501")
|
scripts/generate.py
ADDED
@@ -0,0 +1,167 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# generate.py
|
2 |
+
import sys
|
3 |
+
sys.path.append('versatile_diffusion')
|
4 |
+
import os
|
5 |
+
import os.path as osp
|
6 |
+
import PIL
|
7 |
+
from PIL import Image
|
8 |
+
from pathlib import Path
|
9 |
+
import numpy as np
|
10 |
+
import numpy.random as npr
|
11 |
+
|
12 |
+
import torch
|
13 |
+
import torchvision.transforms as tvtrans
|
14 |
+
from lib.cfg_helper import model_cfg_bank
|
15 |
+
from lib.model_zoo import get_model
|
16 |
+
from lib.model_zoo.ddim_vd import DDIMSampler_VD
|
17 |
+
from lib.experiments.sd_default import color_adjust, auto_merge_imlist
|
18 |
+
from torch.utils.data import DataLoader, Dataset
|
19 |
+
|
20 |
+
from lib.model_zoo.vd import VD
|
21 |
+
from lib.cfg_holder import cfg_unique_holder as cfguh
|
22 |
+
from lib.cfg_helper import get_command_line_args, cfg_initiates, load_cfg_yaml
|
23 |
+
import matplotlib.pyplot as plt
|
24 |
+
from skimage.transform import resize, downscale_local_mean
|
25 |
+
|
26 |
+
|
27 |
+
def regularize_image(x):
|
28 |
+
BICUBIC = PIL.Image.Resampling.BICUBIC
|
29 |
+
if isinstance(x, str):
|
30 |
+
x = Image.open(x).resize([512, 512], resample=BICUBIC)
|
31 |
+
x = tvtrans.ToTensor()(x)
|
32 |
+
elif isinstance(x, PIL.Image.Image):
|
33 |
+
x = x.resize([512, 512], resample=BICUBIC)
|
34 |
+
x = tvtrans.ToTensor()(x)
|
35 |
+
elif isinstance(x, np.ndarray):
|
36 |
+
x = PIL.Image.fromarray(x).resize([512, 512], resample=BICUBIC)
|
37 |
+
x = tvtrans.ToTensor()(x)
|
38 |
+
elif isinstance(x, torch.Tensor):
|
39 |
+
pass
|
40 |
+
else:
|
41 |
+
assert False, 'Unknown image type'
|
42 |
+
|
43 |
+
assert (x.shape[1]==512) & (x.shape[2]==512), \
|
44 |
+
'Wrong image size'
|
45 |
+
return x
|
46 |
+
|
47 |
+
# Load model once globally
|
48 |
+
cfgm_name = 'vd_noema'
|
49 |
+
sampler = DDIMSampler_VD
|
50 |
+
pth = 'versatile_diffusion/pretrained/vd-four-flow-v1-0-fp16-deprecated.pth'
|
51 |
+
cfgm = model_cfg_bank()(cfgm_name)
|
52 |
+
net = get_model()(cfgm)
|
53 |
+
sd = torch.load(pth, map_location='cpu')
|
54 |
+
net.load_state_dict(sd, strict=False)
|
55 |
+
|
56 |
+
# Ensuring proper GPU device assignment, using cuda:0 for all tensor assignments
|
57 |
+
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
|
58 |
+
|
59 |
+
# Move models and data to GPU (cuda:0)
|
60 |
+
net.clip.cuda(0)
|
61 |
+
net.autokl.cuda(0)
|
62 |
+
|
63 |
+
sampler = sampler(net)
|
64 |
+
sampler.model.model.diffusion_model.device = device
|
65 |
+
sampler.model.model.diffusion_model.half().to(device)
|
66 |
+
batch_size = 1
|
67 |
+
|
68 |
+
# Load predicted features and move them to GPU
|
69 |
+
# pred_text = np.load('data/predicted_features/subj{:02d}/nsd_cliptext_predtest_nsdgeneral.npy'.format(sub))
|
70 |
+
# pred_text = torch.tensor(pred_text).half().to(device)
|
71 |
+
|
72 |
+
# pred_vision = np.load('data/predicted_features/subj{:02d}/nsd_clipvision_predtest_nsdgeneral.npy'.format(sub))
|
73 |
+
# pred_vision = torch.tensor(pred_vision).half().to(device)
|
74 |
+
|
75 |
+
n_samples = 1
|
76 |
+
ddim_steps = 50
|
77 |
+
ddim_eta = 0
|
78 |
+
scale = 7.5
|
79 |
+
xtype = 'image'
|
80 |
+
ctype = 'prompt'
|
81 |
+
net.autokl.half()
|
82 |
+
|
83 |
+
torch.manual_seed(0)
|
84 |
+
|
85 |
+
net.clip = net.clip.to(device)
|
86 |
+
|
87 |
+
def generate_image(sub, image_id, annot, strength=0.75, mixing=0.4):
|
88 |
+
|
89 |
+
im_id = image_id
|
90 |
+
|
91 |
+
pred_text = np.load(f'data/predicted_features/subj{sub:02d}/nsd_cliptext_predtest_nsdgeneral.npy')
|
92 |
+
pred_vision = np.load(f'data/predicted_features/subj{sub:02d}/nsd_clipvision_predtest_nsdgeneral.npy')
|
93 |
+
pred_text = torch.tensor(pred_text).half().to(device)
|
94 |
+
pred_vision = torch.tensor(pred_vision).half().to(device)
|
95 |
+
|
96 |
+
zim = Image.open(f'results/vdvae/subj{sub:02d}/{image_id}.png')
|
97 |
+
test_img = Image.open(f'data/nsddata_stimuli/test_images/{image_id}.png')
|
98 |
+
test_img_path = f'scripts/images/original_image.png'
|
99 |
+
test_img.save(test_img_path)
|
100 |
+
|
101 |
+
zim = regularize_image(zim)
|
102 |
+
zin = zim * 2 - 1
|
103 |
+
zin = zin.unsqueeze(0).to(device).half()
|
104 |
+
init_latent = net.autokl_encode(zin)
|
105 |
+
|
106 |
+
sampler.make_schedule(ddim_num_steps=ddim_steps, ddim_eta=ddim_eta, verbose=False)
|
107 |
+
t_enc = int(strength * ddim_steps)
|
108 |
+
z_enc = sampler.stochastic_encode(init_latent, torch.tensor([t_enc]).to(device))
|
109 |
+
|
110 |
+
dummy = ''
|
111 |
+
utx = net.clip_encode_text(dummy).to(device).half()
|
112 |
+
dummy = torch.zeros((1, 3, 224, 224)).to(device)
|
113 |
+
uim = net.clip_encode_vision(dummy).to(device).half()
|
114 |
+
|
115 |
+
z_enc = z_enc.to(device)
|
116 |
+
|
117 |
+
# Sample configuration for diffusion
|
118 |
+
h, w = 512,512
|
119 |
+
shape = [n_samples, 4, h//8, w//8]
|
120 |
+
|
121 |
+
pred_text = np.load(f'data/predicted_features/subj{sub:02d}/nsd_cliptext_predtest_nsdgeneral.npy')
|
122 |
+
with torch.no_grad():
|
123 |
+
pred_text[image_id] = net.clip_encode_text([annot]).to('cpu').numpy().mean(0)
|
124 |
+
pred_text = torch.tensor(pred_text).half().to(device)
|
125 |
+
ctx = pred_text[image_id].unsqueeze(0).to(device)
|
126 |
+
cim = pred_vision[image_id].unsqueeze(0).to(device)
|
127 |
+
|
128 |
+
z = sampler.decode_dc(
|
129 |
+
x_latent=z_enc,
|
130 |
+
first_conditioning=[uim, cim],
|
131 |
+
second_conditioning=[utx, ctx],
|
132 |
+
t_start=t_enc,
|
133 |
+
unconditional_guidance_scale=7.5,
|
134 |
+
xtype='image',
|
135 |
+
first_ctype='vision',
|
136 |
+
second_ctype='prompt',
|
137 |
+
mixed_ratio=(1 - mixing),
|
138 |
+
)
|
139 |
+
|
140 |
+
z = z.to(device).half()
|
141 |
+
x = net.autokl_decode(z)
|
142 |
+
# Adjust color if needed
|
143 |
+
color_adj='None'
|
144 |
+
color_adj_flag = (color_adj != 'none') and (color_adj != 'None') and (color_adj is not None)
|
145 |
+
color_adj_simple = (color_adj == 'Simple') or color_adj == 'simple'
|
146 |
+
color_adj_keep_ratio = 0.5
|
147 |
+
|
148 |
+
if color_adj_flag and (ctype == 'vision'):
|
149 |
+
x_adj = []
|
150 |
+
for xi in x:
|
151 |
+
color_adj_f = color_adjust(ref_from=(xi+1)/2, ref_to=color_adj_to)
|
152 |
+
xi_adj = color_adj_f((xi+1)/2, keep=color_adj_keep_ratio, simple=color_adj_simple)
|
153 |
+
x_adj.append(xi_adj)
|
154 |
+
x = x_adj
|
155 |
+
else:
|
156 |
+
x = torch.clamp((x+1.0)/2.0, min=0.0, max=1.0)
|
157 |
+
x = [tvtrans.ToPILImage()(xi) for xi in x]
|
158 |
+
|
159 |
+
# Save output image
|
160 |
+
x[0].save('scripts/images/reconstructed.png'.format(sub, im_id))
|
161 |
+
# x = torch.clamp((x + 1.0) / 2.0, min=0.0, max=1.0)
|
162 |
+
# x = [tvtrans.ToPILImage()(xi) for xi in x]
|
163 |
+
|
164 |
+
output_path = f'scripts/images/reconstructed.png'
|
165 |
+
# x[0].save(output_path)
|
166 |
+
|
167 |
+
return test_img_path, output_path
|