Spaces:
Running
on
T4
Running
on
T4
File size: 11,975 Bytes
a886fec cbdb77e 06bba7e ff2e0a9 a886fec cd542fa a886fec cd542fa 471a386 06bba7e 53625b9 a886fec 53625b9 a886fec 53625b9 cbdb77e 53625b9 a886fec cbdb77e e99e78e a886fec e99e78e a886fec 06bba7e a886fec cbdb77e a886fec 31070ee 53625b9 31070ee a886fec 53625b9 cd542fa 53625b9 cd542fa 53625b9 cd542fa a886fec cd542fa a886fec 53625b9 31070ee 22e326e 31070ee e628f3f 31070ee e628f3f 31070ee a886fec 22e326e e628f3f 31070ee a886fec cd542fa 31070ee cd542fa e628f3f 31070ee a886fec cd542fa 31070ee cd542fa 31070ee cd542fa cbdb77e cd542fa cbdb77e cd542fa cbdb77e 31070ee a886fec e628f3f a886fec cd542fa 7c9515a 22e326e cd542fa 31070ee 22e326e 31070ee cd542fa 22e326e a886fec 31070ee cd542fa 7c9515a 31070ee cd542fa 31070ee 22e326e 31070ee 22e326e a886fec cd542fa 7c9515a cd542fa 31070ee cd542fa 31070ee a886fec |
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 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 |
import sys
import threading
import streamlit as st
from huggingface_hub import HfFolder, snapshot_download
@st.cache_data
def load_support():
if st.secrets.has_key('etoken'):
HfFolder().save_token(st.secrets['etoken'])
sys.path.append(snapshot_download("OpenShape/openshape-demo-support"))
# st.set_page_config(layout='wide')
load_support()
import numpy
import torch
import openshape
import transformers
from PIL import Image
@st.cache_resource
def load_openshape(name, to_cpu=False):
pce = openshape.load_pc_encoder(name)
if to_cpu:
pce = pce.cpu()
return pce
@st.cache_resource
def load_openclip():
sys.clip_move_lock = threading.Lock()
clip_model, clip_prep = transformers.CLIPModel.from_pretrained(
"laion/CLIP-ViT-bigG-14-laion2B-39B-b160k",
low_cpu_mem_usage=True, torch_dtype=half,
offload_state_dict=True
), transformers.CLIPProcessor.from_pretrained("laion/CLIP-ViT-bigG-14-laion2B-39B-b160k")
if torch.cuda.is_available():
with sys.clip_move_lock:
clip_model.cuda()
return clip_model, clip_prep
f32 = numpy.float32
half = torch.float16 if torch.cuda.is_available() else torch.bfloat16
# clip_model, clip_prep = None, None
clip_model, clip_prep = load_openclip()
model_b32 = load_openshape('openshape-pointbert-vitb32-rgb', True)
model_l14 = load_openshape('openshape-pointbert-vitl14-rgb')
model_g14 = load_openshape('openshape-pointbert-vitg14-rgb')
torch.set_grad_enabled(False)
for kc, vc in st.session_state.get('state_queue', []):
st.session_state[kc] = vc
st.session_state.state_queue = []
import samples_index
from openshape.demo import misc_utils, classification, caption, sd_pc2img, retrieval
st.title("OpenShape Demo")
st.caption("For faster inference without waiting in queue, you may clone the space and run it yourself.")
prog = st.progress(0.0, "Idle")
tab_cls, tab_img, tab_text, tab_pc, tab_sd, tab_cap = st.tabs([
"Classification",
"Retrieval w/ Image",
"Retrieval w/ Text",
"Retrieval w/ 3D",
"Image Generation",
"Captioning",
])
def sq(kc, vc):
st.session_state.state_queue.append((kc, vc))
def reset_3d_shape_input(key):
# this is not working due to streamlit problems, don't use it
model_key = key + "_model"
npy_key = key + "_npy"
swap_key = key + "_swap"
sq(model_key, None)
sq(npy_key, None)
sq(swap_key, "Y is up (for most Objaverse shapes)")
def auto_submit(key):
if st.session_state.get(key):
st.session_state[key] = False
return True
return False
def queue_auto_submit(key):
st.session_state[key] = True
st.experimental_rerun()
img_example_counter = 0
def image_examples(samples, ncols, return_key=None, example_text="Examples"):
global img_example_counter
trigger = False
with st.expander(example_text, True):
for i in range(len(samples) // ncols):
cols = st.columns(ncols)
for j in range(ncols):
idx = i * ncols + j
if idx >= len(samples):
continue
entry = samples[idx]
with cols[j]:
st.image(entry['dispi'])
img_example_counter += 1
with st.columns(5)[2]:
this_trigger = st.button('\+', key='imgexuse%d' % img_example_counter)
trigger = trigger or this_trigger
if this_trigger:
if return_key is None:
for k, v in entry.items():
if not k.startswith('disp'):
sq(k, v)
else:
trigger = entry[return_key]
return trigger
def demo_classification():
with st.form("clsform"):
load_data = misc_utils.input_3d_shape('cls')
cats = st.text_input("Custom Categories (64 max, separated with comma)")
cats = [a.strip() for a in cats.split(',')]
if len(cats) > 64:
st.error('Maximum 64 custom categories supported in the demo')
return
lvis_run = st.form_submit_button("Run Classification on LVIS Categories")
custom_run = st.form_submit_button("Run Classification on Custom Categories")
if lvis_run or auto_submit("clsauto"):
pc = load_data(prog)
col2 = misc_utils.render_pc(pc)
prog.progress(0.5, "Running Classification")
pred = classification.pred_lvis_sims(model_g14, pc)
with col2:
for i, (cat, sim) in zip(range(5), pred.items()):
st.text(cat)
st.caption("Similarity %.4f" % sim)
prog.progress(1.0, "Idle")
if custom_run:
pc = load_data(prog)
col2 = misc_utils.render_pc(pc)
prog.progress(0.5, "Computing Category Embeddings")
device = clip_model.device
tn = clip_prep(text=cats, return_tensors='pt', truncation=True, max_length=76).to(device)
feats = clip_model.get_text_features(**tn).float().cpu()
prog.progress(0.5, "Running Classification")
pred = classification.pred_custom_sims(model_g14, pc, cats, feats)
with col2:
for i, (cat, sim) in zip(range(5), pred.items()):
st.text(cat)
st.caption("Similarity %.4f" % sim)
prog.progress(1.0, "Idle")
if image_examples(samples_index.classification, 3, example_text="Examples (Choose one of the following 3D shapes)"):
queue_auto_submit("clsauto")
def demo_captioning():
with st.form("capform"):
load_data = misc_utils.input_3d_shape('cap')
cond_scale = st.slider('Conditioning Scale', 0.0, 4.0, 2.0, 0.1, key='capcondscl')
if st.form_submit_button("Generate a Caption") or auto_submit("capauto"):
pc = load_data(prog)
col2 = misc_utils.render_pc(pc)
prog.progress(0.5, "Running Generation")
cap = caption.pc_caption(model_b32, pc, cond_scale)
st.text(cap)
prog.progress(1.0, "Idle")
if image_examples(samples_index.cap, 3, example_text="Examples (Choose one of the following 3D shapes)"):
queue_auto_submit("capauto")
def demo_pc2img():
with st.form("sdform"):
load_data = misc_utils.input_3d_shape('sd')
prompt = st.text_input("Prompt (Optional)", key='sdtprompt')
noise_scale = st.slider('Variation Level', 0, 5, 1)
cfg_scale = st.slider('Guidance Scale', 0.0, 30.0, 10.0)
steps = st.slider('Diffusion Steps', 8, 50, 25)
width = 640 # st.slider('Width', 480, 640, step=32)
height = 640 # st.slider('Height', 480, 640, step=32)
if st.form_submit_button("Generate") or auto_submit("sdauto"):
pc = load_data(prog)
col2 = misc_utils.render_pc(pc)
prog.progress(0.49, "Running Generation")
if torch.cuda.is_available():
with sys.clip_move_lock:
clip_model.cpu()
img = sd_pc2img.pc_to_image(
model_l14, pc, prompt, noise_scale, width, height, cfg_scale, steps,
lambda i, t, _: prog.progress(0.49 + i / (steps + 1) / 2, "Running Diffusion Step %d" % i)
)
if torch.cuda.is_available():
with sys.clip_move_lock:
clip_model.cuda()
with col2:
st.image(img)
prog.progress(1.0, "Idle")
if image_examples(samples_index.sd, 3, example_text="Examples (Choose one of the following 3D shapes)"):
queue_auto_submit("sdauto")
def retrieval_results(results):
st.caption("Click the link to view the 3D shape")
for i in range(len(results) // 4):
cols = st.columns(4)
for j in range(4):
idx = i * 4 + j
if idx >= len(results):
continue
entry = results[idx]
with cols[j]:
ext_link = f"https://objaverse.allenai.org/explore/?query={entry['u']}"
st.image(entry['img'])
# st.markdown(f"[![thumbnail {entry['desc'].replace('\n', ' ')}]({entry['img']})]({ext_link})")
# st.text(entry['name'])
quote_name = entry['name'].replace('[', '\\[').replace(']', '\\]').replace('\n', ' ')
st.markdown(f"[{quote_name}]({ext_link})")
def demo_retrieval():
with tab_text:
with st.form("rtextform"):
k = st.slider("Shapes to Retrieve", 1, 100, 16, key='rtext')
text = st.text_input("Input Text", key="inputrtext")
if st.form_submit_button("Run with Text") or auto_submit("rtextauto"):
prog.progress(0.49, "Computing Embeddings")
device = clip_model.device
tn = clip_prep(
text=[text], return_tensors='pt', truncation=True, max_length=76
).to(device)
enc = clip_model.get_text_features(**tn).float().cpu()
prog.progress(0.7, "Running Retrieval")
retrieval_results(retrieval.retrieve(enc, k))
prog.progress(1.0, "Idle")
picked_sample = st.selectbox("Examples", ["Select..."] + samples_index.retrieval_texts)
text_last_example = st.session_state.get('text_last_example', None)
if text_last_example is None:
st.session_state.text_last_example = picked_sample
elif text_last_example != picked_sample and picked_sample != "Select...":
st.session_state.text_last_example = picked_sample
sq("inputrtext", picked_sample)
queue_auto_submit("rtextauto")
with tab_img:
submit = False
with st.form("rimgform"):
k = st.slider("Shapes to Retrieve", 1, 100, 16, key='rimage')
pic = st.file_uploader("Upload an Image", key='rimageinput')
if st.form_submit_button("Run with Image"):
submit = True
results_container = st.container()
sample_got = image_examples(samples_index.iret, 4, 'rimageinput')
if sample_got:
pic = sample_got
if sample_got or submit:
img = Image.open(pic)
with results_container:
st.image(img)
prog.progress(0.49, "Computing Embeddings")
device = clip_model.device
tn = clip_prep(images=[img], return_tensors="pt").to(device)
enc = clip_model.get_image_features(pixel_values=tn['pixel_values'].type(half)).float().cpu()
prog.progress(0.7, "Running Retrieval")
retrieval_results(retrieval.retrieve(enc, k))
prog.progress(1.0, "Idle")
with tab_pc:
with st.form("rpcform"):
k = st.slider("Shapes to Retrieve", 1, 100, 16, key='rpc')
load_data = misc_utils.input_3d_shape('retpc')
if st.form_submit_button("Run with Shape") or auto_submit('rpcauto'):
pc = load_data(prog)
col2 = misc_utils.render_pc(pc)
prog.progress(0.49, "Computing Embeddings")
ref_dev = next(model_g14.parameters()).device
enc = model_g14(torch.tensor(pc[:, [0, 2, 1, 3, 4, 5]].T[None], device=ref_dev)).cpu()
prog.progress(0.7, "Running Retrieval")
retrieval_results(retrieval.retrieve(enc, k))
prog.progress(1.0, "Idle")
if image_examples(samples_index.pret, 3):
queue_auto_submit("rpcauto")
try:
with tab_cls:
demo_classification()
with tab_cap:
demo_captioning()
with tab_sd:
demo_pc2img()
demo_retrieval()
except Exception:
import traceback
st.error(traceback.format_exc().replace("\n", " \n"))
|