Spaces:
Sleeping
Sleeping
File size: 17,005 Bytes
823b132 1c79692 823b132 38b8e35 823b132 38b8e35 823b132 38b8e35 823b132 1c79692 38b8e35 823b132 38b8e35 823b132 1c79692 38b8e35 1c79692 38b8e35 1c79692 38b8e35 1c79692 3b1707b | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 | import gradio as gr
import json, time, uuid, os, csv, random
import numpy as np
TRIALS_JSON = "trials.json"
LOG_PATH = "responses.csv"
# Check for persistent storage (Hugging Face Spaces)
if os.path.exists("/data"):
LOG_PATH = "/data/responses.csv"
def load_trials(path=TRIALS_JSON):
with open(path, "r") as f:
return json.load(f)
TRIALS = load_trials()
def ensure_log_header():
if not os.path.exists(LOG_PATH):
with open(LOG_PATH, "w", newline="") as f:
w = csv.writer(f)
w.writerow([
"ts", "participant_id", "trial_id", "task",
"left_method", "right_method",
"choice", "voxel_choice", "rt_ms",
"prompt_text", "prompt_image"
])
ensure_log_header()
import os, hashlib
import trimesh
UNCOLORED_DIR = "cache_uncolored"
os.makedirs(UNCOLORED_DIR, exist_ok=True)
def to_uncolored_obj(mesh_path: str) -> str:
# cache by path string (and you can include file mtime if you want)
key = hashlib.md5(mesh_path.encode("utf-8")).hexdigest()
out_path = os.path.join(UNCOLORED_DIR, f"{key}.glb")
if os.path.exists(out_path):
return out_path
m = trimesh.load(mesh_path, force="mesh", process=False)
# --- THIS is the key line: only verts + faces, no visual/material info ---
# Preserving normals is crucial for correct shading
m = trimesh.Trimesh(
vertices=m.vertices,
faces=m.faces,
# vertex_normals=m.vertex_normals,
process=False
)
# export as OBJ (no textures/colors)
m.export(out_path)
return out_path
def to_gray_obj(mesh_path: str, gray=180) -> str:
"""
Convert any mesh (obj/glb/gltf/...) to a geometry-only OBJ with constant gray vertex colors.
gray: 0..255
"""
# Updated cache key to 'v4_double_sided' to force regeneration
key = hashlib.md5((mesh_path + f":g{gray}:v4_double_sided").encode("utf-8")).hexdigest()
out_path = os.path.join(UNCOLORED_DIR, f"{key}.glb")
if os.path.exists(out_path):
return out_path
# process=True fixes winding, merges vertices, etc.
loaded = trimesh.load(mesh_path, process=True)
# GLB/GLTF often loads as a Scene -> merge geometries
if isinstance(loaded, trimesh.Scene):
if len(loaded.geometry) == 0:
raise ValueError(f"Empty scene: {mesh_path}")
m = trimesh.util.concatenate(tuple(loaded.geometry.values()))
else:
m = loaded
# Fix normals to ensure smooth shading (computes them if missing)
# m.fix_normals()
# Constant vertex colors (RGBA)
vc = np.tile(np.array([gray, gray, gray, 255], dtype=np.uint8), (len(m.vertices), 1))
m.visual = trimesh.visual.ColorVisuals(mesh=m, vertex_colors=vc)
# Enable double-sided rendering to fix visibility issues
# if hasattr(m.visual, 'material'):
# m.visual.material.doubleSided = True
# For some Trimesh versions/types, we might need to ensure it's PBR or handled correctly
# But usually setting the attribute on the material object helps GLB export.
# Export as GLB (supports vertex colors + normals well)
m.export(out_path)
return out_path
def to_ccm_obj(mesh_path: str, gray=180) -> str:
"""
Convert any mesh (obj/glb/gltf/...) to a geometry-only OBJ with constant gray vertex colors.
gray: 0..255
"""
# Updated cache key to 'v4_double_sided' to force regeneration
key = hashlib.md5((mesh_path + f":g{gray}:v4_ccm").encode("utf-8")).hexdigest()
out_path = os.path.join(UNCOLORED_DIR, f"{key}.glb")
if os.path.exists(out_path):
return out_path
# process=True fixes winding, merges vertices, etc.
loaded = trimesh.load(mesh_path, process=True)
# GLB/GLTF often loads as a Scene -> merge geometries
if isinstance(loaded, trimesh.Scene):
if len(loaded.geometry) == 0:
raise ValueError(f"Empty scene: {mesh_path}")
m = trimesh.util.concatenate(tuple(loaded.geometry.values()))
else:
m = loaded
# Fix normals to ensure smooth shading (computes them if missing)
# m.fix_normals()
# Constant vertex colors (RGBA)
# vc = np.tile(np.array([gray, gray, gray, 255], dtype=np.uint8), (len(m.vertices), 1))
num_vertices = len(m.vertices)
# color based on coord
vc = m.vertices.copy()
vc = (vc - vc.min(axis=0)) / (vc.max(axis=0) - vc.min(axis=0))
vc = (vc * 255).astype(np.uint8)
vc_alpha = np.ones((num_vertices, 1), dtype=np.uint8) * 255
vc = np.concatenate([vc, vc_alpha], axis=1)
# normalize vertices to 0,1
m.visual = trimesh.visual.ColorVisuals(mesh=m, vertex_colors=vc)
# Enable double-sided rendering to fix visibility issues
# if hasattr(m.visual, 'material'):
# m.visual.material.doubleSided = True
# For some Trimesh versions/types, we might need to ensure it's PBR or handled correctly
# But usually setting the attribute on the material object helps GLB export.
# Export as GLB (supports vertex colors + normals well)
m.export(out_path)
return out_path
def to_normal_obj(mesh_path: str, gray=180) -> str:
"""
Convert any mesh (obj/glb/gltf/...) to a geometry-only OBJ with constant gray vertex colors.
gray: 0..255
"""
# Updated cache key to 'v4_double_sided' to force regeneration
key = hashlib.md5((mesh_path + f":g{gray}:v4_normal").encode("utf-8")).hexdigest()
out_path = os.path.join(UNCOLORED_DIR, f"{key}.glb")
if os.path.exists(out_path):
return out_path
# process=True fixes winding, merges vertices, etc.
loaded = trimesh.load(mesh_path, process=True)
# GLB/GLTF often loads as a Scene -> merge geometries
if isinstance(loaded, trimesh.Scene):
if len(loaded.geometry) == 0:
raise ValueError(f"Empty scene: {mesh_path}")
m = trimesh.util.concatenate(tuple(loaded.geometry.values()))
else:
m = loaded
# Fix normals to ensure smooth shading (computes them if missing)
m.fix_normals()
# Normalize normals to 0..1 range for visualization
# Normal range is [-1, 1], so (n + 1) / 2 maps to [0, 1]
vc = (m.vertex_normals + 1) / 2
vc = (vc * 255).astype(np.uint8)
# Add alpha channel
num_vertices = len(m.vertices)
vc_alpha = np.ones((num_vertices, 1), dtype=np.uint8) * 255
vc = np.concatenate([vc, vc_alpha], axis=1)
m.visual = trimesh.visual.ColorVisuals(mesh=m, vertex_colors=vc)
# Enable double-sided rendering to fix visibility issues
# if hasattr(m.visual, 'material'):
# m.visual.material.doubleSided = True
# For some Trimesh versions/types, we might need to ensure it's PBR or handled correctly
# But usually setting the attribute on the material object helps GLB export.
# Export as GLB (supports vertex colors + normals well)
m.export(out_path)
return out_path
def _trial_to_view(trial, flip_lr: bool):
if flip_lr:
left_method, right_method = "continuous", "ours"
left_voxel, right_voxel = trial["cont_voxel"], trial["ours_voxel"]
left_mesh, right_mesh = trial["cont_mesh"], trial["ours_mesh"]
left_video = trial.get("cont_video", None)
right_video = trial.get("ours_video", None)
else:
left_method, right_method = "ours", "continuous"
left_voxel, right_voxel = trial["ours_voxel"], trial["cont_voxel"]
left_mesh, right_mesh = trial["ours_mesh"], trial["cont_mesh"]
left_video = trial.get("ours_video", None)
right_video = trial.get("cont_video", None)
# Make sure voxel OBJ is geometry-only (and optionally also strip mesh if you want)
left_voxel = to_ccm_obj(left_voxel)
right_voxel = to_ccm_obj(right_voxel)
# (Optional) also strip mesh to geometry-only if needed:
left_mesh = to_normal_obj(left_mesh) # only if you want to force mesh untextured via OBJ export
right_mesh = to_normal_obj(right_mesh)
task = trial["task"]
prompt_text = trial.get("prompt_text", "")
prompt_image = trial.get("prompt_image", None)
show_text = (task == "text")
show_img = (task == "image")
return dict(
trial_id=trial["trial_id"],
task=task,
show_text=show_text,
show_img=show_img,
prompt_text=prompt_text,
prompt_image=prompt_image,
left_method=left_method,
right_method=right_method,
left_voxel=left_voxel,
right_voxel=right_voxel,
left_mesh=left_mesh,
right_mesh=right_mesh,
left_video=left_video,
right_video=right_video,
)
def start_session(seed=None):
pid = str(uuid.uuid4())[:8]
rng = random.Random(seed if seed is not None else time.time())
order = list(range(len(TRIALS)))
rng.shuffle(order)
# DEBUG: move specific trial to front
# target_id = "img_treehouse_rmapple"
# for i, t in enumerate(TRIALS):
# if t["trial_id"] == target_id:
# if i in order:
# order.remove(i)
# order.insert(0, i)
# break
state = {
"pid": pid,
"order": order,
"i": 0,
"t0": time.time(),
"flip_lr": rng.choice([False, True]), # for first trial
"rng_state": rng.getstate(),
}
view = _trial_to_view(TRIALS[order[0]], state["flip_lr"])
return (
state, pid,
gr.update(value=f"Trial 1 / {len(TRIALS)}"),
gr.update(visible=view["show_text"], value=view["prompt_text"]),
gr.update(visible=view["show_img"], value=view["prompt_image"]),
gr.update(value="Left Model"),
gr.update(value="Right Model"),
view["left_voxel"], view["right_voxel"],
view["left_mesh"], view["right_mesh"],
view["left_video"], view["right_video"],
gr.update(value=""), # completion code
gr.update(value=None), # reset choice
gr.update(value=None) # reset voxel_choice
)
def submit(choice, voxel_choice, state):
i = state["i"]
idx = state["order"][i]
trial = TRIALS[idx]
rt_ms = int((time.time() - state["t0"]) * 1000)
# log
with open(LOG_PATH, "a", newline="") as f:
w = csv.writer(f)
w.writerow([
time.time(), state["pid"], trial["trial_id"], trial["task"],
("continuous" if state["flip_lr"] else "ours"),
("ours" if state["flip_lr"] else "continuous"),
choice, voxel_choice, rt_ms,
trial.get("prompt_text", ""),
trial.get("prompt_image", ""),
])
# advance
state["i"] += 1
if state["i"] >= len(TRIALS):
code = f"COMPLETE-{state['pid']}"
return (
state,
gr.update(value="Done ✅ (copy the completion code below)"),
gr.update(visible=False, value=""),
gr.update(visible=False, value=None),
gr.update(value="LEFT"),
gr.update(value="RIGHT"),
None, None,
None, None,
None, None,
gr.update(value=code),
gr.update(value=None),
gr.update(value=None)
)
# next trial
rng = random.Random()
rng.setstate(state["rng_state"])
state["flip_lr"] = rng.choice([False, True])
state["rng_state"] = rng.getstate()
state["t0"] = time.time()
ni = state["i"]
nidx = state["order"][ni]
view = _trial_to_view(TRIALS[nidx], state["flip_lr"])
return (
state,
gr.update(value=f"Trial {ni+1} / {len(TRIALS)}"),
gr.update(visible=view["show_text"], value=view["prompt_text"]),
gr.update(visible=view["show_img"], value=view["prompt_image"]),
gr.update(value="Left Model"),
gr.update(value="Right Model"),
view["left_voxel"], view["right_voxel"],
view["left_mesh"], view["right_mesh"],
view["left_video"], view["right_video"],
gr.update(value=""),
gr.update(value=None),
gr.update(value=None)
)
def get_responses_file(password):
# Set your password in HF Space Settings as ADMIN_PASSWORD, or use default "123456"
correct_pass = os.environ.get("ADMIN_PASSWORD", "123456")
if password != correct_pass:
raise gr.Error("Incorrect Password")
if not os.path.exists(LOG_PATH):
raise gr.Error("No responses collected yet.")
return LOG_PATH
def clear_responses(password):
correct_pass = os.environ.get("ADMIN_PASSWORD", "123456")
if password != correct_pass:
raise gr.Error("Incorrect Password")
if os.path.exists(LOG_PATH):
os.remove(LOG_PATH)
ensure_log_header() # Re-create header immediately
return "Responses cleared successfully."
else:
return "No responses file found to clear."
with gr.Blocks(css="""
#prompt-img {max-width: 320px; margin: 0 auto;}
#prompt-img img {width: 100% !important; height: auto !important;}
.model3d-window {height: 600px !important;}
""") as demo:
st = gr.State()
gr.Markdown("## 3D Asset User Study (Image + Text)")
with gr.Accordion("📝 Instructions (Click to Read)", open=True):
gr.Markdown("""
### How to complete this study:
1. **Observe the Prompt**: You will see a text description or a reference image.
2. **Compare Models**: Rotate and zoom into the "Left" and "Right" 3D models to inspect them.
3. **Vote**: Select which model, voxel (Geometry and/or Texture) better matches the prompt or has higher quality.
4. **Submit**: Click "Submit" to move to the next trial.
*You could consider:
1. Alignment with prompt.
2. Stuctural quality, such as holes, cracks, floaters, etc.
3. Since the process of voxel->mesh can be lossy (missing parts), please also inspect voxels' quality.
4. If you are not sure about the winner, pick not sure.
5. Unexpected holes on voxels are typically not desired.*
""")
pid_box = gr.Textbox(label="Participant ID (auto)", interactive=False)
progress = gr.Markdown()
# Prompt area
prompt_text = gr.Markdown(visible=False)
prompt_img = gr.Image(label="Prompt Image", visible=False, elem_id="prompt-img")
gr.Markdown("### Compare the two methods")
with gr.Row():
with gr.Column():
left_title = gr.Textbox(label="Left Method", interactive=False)
left_voxel = gr.Model3D(label="Left Voxel (untextured)", clear_color=[1,1,1,1], elem_classes="model3d-window")
left_mesh = gr.Model3D(label="Left Mesh (untextured)", clear_color=[1,1,1,1], elem_classes="model3d-window")
left_vid = gr.Video(label="Left Video (optional)", autoplay=True)
with gr.Column():
right_title = gr.Textbox(label="Right Method", interactive=False)
right_voxel = gr.Model3D(label="Right Voxel (untextured)", clear_color=[1,1,1,1], elem_classes="model3d-window")
right_mesh = gr.Model3D(label="Right Mesh (untextured)", clear_color=[1,1,1,1], elem_classes="model3d-window")
right_vid = gr.Video(label="Right Video (optional)", autoplay=True)
voxel_choice = gr.Radio(["Left", "Right", "Not Sure"], label="Which generated voxel is better?", value=None)
choice = gr.Radio(["Left", "Right", "Not Sure"], label="Which generated mesh is better?", value=None)
btn = gr.Button("Submit")
code_box = gr.Textbox(label="Completion Code", interactive=False)
demo.load(
start_session,
inputs=None,
outputs=[st, pid_box, progress, prompt_text, prompt_img,
left_title, right_title,
left_voxel, right_voxel, left_mesh, right_mesh, left_vid, right_vid,
code_box, choice, voxel_choice]
)
btn.click(
submit,
inputs=[choice, voxel_choice, st],
outputs=[st, progress, prompt_text, prompt_img,
left_title, right_title,
left_voxel, right_voxel, left_mesh, right_mesh, left_vid, right_vid,
code_box, choice, voxel_choice]
)
gr.Markdown("---")
with gr.Accordion("Admin Zone (Download Responses)", open=False):
with gr.Row():
admin_pass = gr.Textbox(label="Admin Password", type="password", placeholder="Enter password to download")
download_btn = gr.Button("Download CSV")
clear_btn = gr.Button("Clear Responses", variant="stop")
with gr.Row():
admin_file = gr.File(label="Responses File", interactive=False)
admin_status = gr.Textbox(label="Status", interactive=False)
download_btn.click(
get_responses_file,
inputs=[admin_pass],
outputs=[admin_file]
)
clear_btn.click(
clear_responses,
inputs=[admin_pass],
outputs=[admin_status]
)
demo.launch(allowed_paths=["/data"])
|