Spaces:
Running
on
Zero
Running
on
Zero
add one click generate glb
Browse files
app.py
CHANGED
@@ -119,7 +119,7 @@ def image_to_3d(
|
|
119 |
slat_sampling_steps: int,
|
120 |
multiimage_algo: Literal["multidiffusion", "stochastic"],
|
121 |
req: gr.Request,
|
122 |
-
) -> Tuple[dict, str
|
123 |
"""
|
124 |
Convert an image (or multiple images) into a 3D model and return its state and video.
|
125 |
|
@@ -137,7 +137,6 @@ def image_to_3d(
|
|
137 |
Returns:
|
138 |
dict: The information of the generated 3D model.
|
139 |
str: The path to the video of the 3D model.
|
140 |
-
str: serialized JSON of state
|
141 |
|
142 |
"""
|
143 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
@@ -188,7 +187,7 @@ def image_to_3d(
|
|
188 |
# Pack state for downstream use
|
189 |
state = pack_state(outputs['gaussian'][0], outputs['mesh'][0])
|
190 |
torch.cuda.empty_cache()
|
191 |
-
return state, video_path
|
192 |
|
193 |
|
194 |
|
@@ -321,11 +320,15 @@ with gr.Blocks(delete_cache=(600, 600)) as demo:
|
|
321 |
|
322 |
with gr.Row():
|
323 |
download_glb = gr.DownloadButton(label="Download GLB", interactive=False)
|
324 |
-
download_gs = gr.DownloadButton(label="Download Gaussian", interactive=False)
|
|
|
|
|
|
|
|
|
|
|
325 |
|
326 |
is_multiimage = gr.State(False)
|
327 |
output_buf = gr.State()
|
328 |
-
state_textbox = gr.Textbox(visible=False, label="Serialized State")
|
329 |
|
330 |
# Example images at the bottom of the page
|
331 |
with gr.Row() as single_image_example:
|
@@ -385,7 +388,7 @@ with gr.Blocks(delete_cache=(600, 600)) as demo:
|
|
385 |
ss_guidance_strength, ss_sampling_steps,
|
386 |
slat_guidance_strength, slat_sampling_steps, multiimage_algo
|
387 |
],
|
388 |
-
outputs=[output_buf, video_output
|
389 |
).then(
|
390 |
lambda: tuple([gr.Button(interactive=True), gr.Button(interactive=True)]),
|
391 |
outputs=[extract_glb_btn, extract_gs_btn],
|
@@ -418,6 +421,27 @@ with gr.Blocks(delete_cache=(600, 600)) as demo:
|
|
418 |
lambda: gr.Button(interactive=False),
|
419 |
outputs=[download_glb],
|
420 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
421 |
|
422 |
|
423 |
# Launch the Gradio app
|
|
|
119 |
slat_sampling_steps: int,
|
120 |
multiimage_algo: Literal["multidiffusion", "stochastic"],
|
121 |
req: gr.Request,
|
122 |
+
) -> Tuple[dict, str]:
|
123 |
"""
|
124 |
Convert an image (or multiple images) into a 3D model and return its state and video.
|
125 |
|
|
|
137 |
Returns:
|
138 |
dict: The information of the generated 3D model.
|
139 |
str: The path to the video of the 3D model.
|
|
|
140 |
|
141 |
"""
|
142 |
user_dir = os.path.join(TMP_DIR, str(req.session_hash))
|
|
|
187 |
# Pack state for downstream use
|
188 |
state = pack_state(outputs['gaussian'][0], outputs['mesh'][0])
|
189 |
torch.cuda.empty_cache()
|
190 |
+
return state, video_path
|
191 |
|
192 |
|
193 |
|
|
|
320 |
|
321 |
with gr.Row():
|
322 |
download_glb = gr.DownloadButton(label="Download GLB", interactive=False)
|
323 |
+
download_gs = gr.DownloadButton(label="Download Gaussian", interactive=False)
|
324 |
+
|
325 |
+
with gr.Accordion("Quick GLB from Image", open=False):
|
326 |
+
generate_glb_btn = gr.Button("Upload and Generate GLB Automatically")
|
327 |
+
quick_video = gr.Video(label="Quick 3D Preview", autoplay=True, loop=True)
|
328 |
+
quick_glb_download = gr.DownloadButton(label="Download GLB", interactive=False)
|
329 |
|
330 |
is_multiimage = gr.State(False)
|
331 |
output_buf = gr.State()
|
|
|
332 |
|
333 |
# Example images at the bottom of the page
|
334 |
with gr.Row() as single_image_example:
|
|
|
388 |
ss_guidance_strength, ss_sampling_steps,
|
389 |
slat_guidance_strength, slat_sampling_steps, multiimage_algo
|
390 |
],
|
391 |
+
outputs=[output_buf, video_output],
|
392 |
).then(
|
393 |
lambda: tuple([gr.Button(interactive=True), gr.Button(interactive=True)]),
|
394 |
outputs=[extract_glb_btn, extract_gs_btn],
|
|
|
421 |
lambda: gr.Button(interactive=False),
|
422 |
outputs=[download_glb],
|
423 |
)
|
424 |
+
|
425 |
+
generate_glb_btn.click(
|
426 |
+
lambda: get_seed(True, 0),
|
427 |
+
outputs=[seed]
|
428 |
+
).then(
|
429 |
+
image_to_3d,
|
430 |
+
inputs=[
|
431 |
+
image_prompt,
|
432 |
+
gr.State([]),
|
433 |
+
gr.State(False),
|
434 |
+
seed,
|
435 |
+
gr.State(7.5), gr.State(12),
|
436 |
+
gr.State(3.0), gr.State(12),
|
437 |
+
gr.State("stochastic")
|
438 |
+
],
|
439 |
+
outputs=[output_buf, quick_video],
|
440 |
+
).then(
|
441 |
+
extract_glb,
|
442 |
+
inputs=[output_buf, mesh_simplify, texture_size],
|
443 |
+
outputs=[model_output, quick_glb_download]
|
444 |
+
)
|
445 |
|
446 |
|
447 |
# Launch the Gradio app
|