Update app.py

optimisation and timeout issues fixed?

app.py

@@ -1,44 +1,4 @@
-import gradio as gr
-import spaces
-from gradio_litmodel3d import LitModel3D
-
-import os
-import shutil
-os.environ['SPCONV_ALGO'] = 'native'
-from typing import *
-import torch
-import numpy as np
-import imageio
-import uuid
-from easydict import EasyDict as edict
-from PIL import Image
-from trellis.pipelines import TrellisImageTo3DPipeline
-from trellis.representations import Gaussian, MeshExtractResult
-from trellis.utils import render_utils, postprocessing_utils
-
-
-MAX_SEED = np.iinfo(np.int32).max
-TMP_DIR = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'tmp')
-os.makedirs(TMP_DIR, exist_ok=True)
-
-
-def start_session(req: gr.Request):
-    user_dir = os.path.join(TMP_DIR, str(req.session_hash))
-    print(f'Creating user directory: {user_dir}')
-    os.makedirs(user_dir, exist_ok=True)
-    
-def end_session(req: gr.Request):
-    user_dir = os.path.join(TMP_DIR, str(req.session_hash))
-    print(f'Removing user directory: {user_dir}')
-    shutil.rmtree(user_dir)
-
-def preprocess_image(image: Image.Image) -> Tuple[str, Image.Image]:
-    processed_image = pipeline.preprocess_image(image)
-    return processed_image
-
-def get_seed(randomize_seed: bool, seed: int) -> int:
-    """Get the random seed."""
-    return np.random.randint(0, MAX_SEED) if randomize_seed else seed
+# ... (previous imports remain the same) ...
 
 @spaces.GPU
 def image_to_3d(
@@ -49,125 +9,162 @@ def image_to_3d(
     slat_guidance_strength: float,
     slat_sampling_steps: int,
     req: gr.Request,
-) -> Tuple[str, str, str]:
+    progress: gr.Progress = gr.Progress()
+) -> Tuple[dict, str, str, str]:
     """
-    Convert an image to a 3D model and save both video preview and full-quality GLB.
-    
-    Returns:
-        Tuple[str, str, str]: (video_path, glb_path, download_path)
+    Convert an image to a 3D model with improved memory management and progress tracking.
     """
     user_dir = os.path.join(TMP_DIR, str(req.session_hash))
-    outputs = pipeline.run(
-        image,
-        seed=seed,
-        formats=["gaussian", "mesh"],
-        preprocess_image=False,
-        sparse_structure_sampler_params={
-            "steps": ss_sampling_steps,
-            "cfg_strength": ss_guidance_strength,
-        },
-        slat_sampler_params={
-            "steps": slat_sampling_steps,
-            "cfg_strength": slat_guidance_strength,
-        },
-    )
-    
-    # Generate and save video preview
-    video = render_utils.render_video(outputs['gaussian'][0], num_frames=120)['color']
-    video_geo = render_utils.render_video(outputs['mesh'][0], num_frames=120)['normal']
-    video = [np.concatenate([video[i], video_geo[i]], axis=1) for i in range(len(video))]
-    trial_id = str(uuid.uuid4())
-    video_path = os.path.join(user_dir, f"{trial_id}.mp4")
-    imageio.mimsave(video_path, video, fps=15)
-    
-    # Save full-quality GLB directly from the generated mesh
-    glb = postprocessing_utils.to_glb(
-        outputs['gaussian'][0], 
-        outputs['mesh'][0],
-        simplify=0.0,  # No simplification
-        texture_size=2048,  # Maximum texture resolution
-        verbose=False
-    )
-    glb_path = os.path.join(user_dir, f"{trial_id}_full.glb")
-    glb.export(glb_path)
+    progress(0, desc="Initializing...")
     
+    # Clear CUDA cache before starting
     torch.cuda.empty_cache()
-    return video_path, glb_path, glb_path
-
-with gr.Blocks(delete_cache=(600, 600)) as demo:
-    gr.Markdown("""
-    ## Image to 3D Asset with [TRELLIS](https://trellis3d.github.io/)
-    * Upload an image and click "Generate" to create a high-quality 3D model
-    * Once generation is complete, you can download the full-quality GLB file
-    * The model will be in maximum quality with no reduction applied
-    """)
     
-    with gr.Row():
-        with gr.Column():
-            image_prompt = gr.Image(label="Image Prompt", format="png", image_mode="RGBA", type="pil", height=300)
+    try:
+        # Generate 3D model with progress updates
+        progress(0.1, desc="Running 3D generation pipeline...")
+        outputs = pipeline.run(
+            image,
+            seed=seed,
+            formats=["gaussian", "mesh"],
+            preprocess_image=False,
+            sparse_structure_sampler_params={
+                "steps": ss_sampling_steps,
+                "cfg_strength": ss_guidance_strength,
+            },
+            slat_sampler_params={
+                "steps": slat_sampling_steps,
+                "cfg_strength": slat_guidance_strength,
+            },
+        )
+        
+        progress(0.4, desc="Generating video preview...")
+        # Generate video frames in batches to manage memory
+        batch_size = 30  # Process 30 frames at a time
+        num_frames = 120
+        video = []
+        video_geo = []
+        
+        for i in range(0, num_frames, batch_size):
+            end_idx = min(i + batch_size, num_frames)
+            batch_frames = render_utils.render_video(
+                outputs['gaussian'][0], 
+                num_frames=end_idx - i, 
+                start_frame=i
+            )['color']
+            batch_geo = render_utils.render_video(
+                outputs['mesh'][0], 
+                num_frames=end_idx - i,
+                start_frame=i
+            )['normal']
             
-            with gr.Accordion(label="Generation Settings", open=False):
-                seed = gr.Slider(0, MAX_SEED, label="Seed", value=0, step=1)
-                randomize_seed = gr.Checkbox(label="Randomize Seed", value=True)
-                gr.Markdown("Stage 1: Sparse Structure Generation")
-                with gr.Row():
-                    ss_guidance_strength = gr.Slider(0.0, 10.0, label="Guidance Strength", value=7.5, step=0.1)
-                    ss_sampling_steps = gr.Slider(1, 500, label="Sampling Steps", value=12, step=1)
-                gr.Markdown("Stage 2: Structured Latent Generation")
-                with gr.Row():
-                    slat_guidance_strength = gr.Slider(0.0, 10.0, label="Guidance Strength", value=3.0, step=0.1)
-                    slat_sampling_steps = gr.Slider(1, 500, label="Sampling Steps", value=12, step=1)
-
-            generate_btn = gr.Button("Generate")
-
-        with gr.Column():
-            video_output = gr.Video(label="Generated 3D Asset Preview", autoplay=True, loop=True, height=300)
-            model_output = LitModel3D(label="3D Model Preview", exposure=20.0, height=300)
-            download_glb = gr.DownloadButton(label="Download Full-Quality GLB", interactive=False)
+            video.extend(batch_frames)
+            video_geo.extend(batch_geo)
             
-    # Example images
-    with gr.Row():
-        examples = gr.Examples(
-            examples=[
-                f'assets/example_image/{image}'
-                for image in os.listdir("assets/example_image")
-            ],
-            inputs=[image_prompt],
-            fn=preprocess_image,
-            outputs=[image_prompt],
-            run_on_click=True,
-            examples_per_page=64,
+            # Clear cache after each batch
+            torch.cuda.empty_cache()
+            progress(0.4 + (0.3 * i / num_frames), desc=f"Rendering frames {i} to {end_idx}...")
+        
+        # Combine video frames
+        video = [np.concatenate([video[i], video_geo[i]], axis=1) for i in range(len(video))]
+        
+        # Generate unique ID and save video
+        trial_id = str(uuid.uuid4())
+        video_path = os.path.join(user_dir, f"{trial_id}.mp4")
+        progress(0.7, desc="Saving video...")
+        imageio.mimsave(video_path, video, fps=15)
+        
+        # Clear video data from memory
+        del video
+        del video_geo
+        torch.cuda.empty_cache()
+        
+        # Generate and save full-quality GLB
+        progress(0.8, desc="Generating full-quality GLB...")
+        glb = postprocessing_utils.to_glb(
+            outputs['gaussian'][0], 
+            outputs['mesh'][0],
+            simplify=0.0,
+            texture_size=2048,
+            verbose=False
         )
+        glb_path = os.path.join(user_dir, f"{trial_id}_full.glb")
+        progress(0.9, desc="Saving GLB file...")
+        glb.export(glb_path)
+        
+        # Pack state for reduced version
+        progress(0.95, desc="Finalizing...")
+        state = pack_state(outputs['gaussian'][0], outputs['mesh'][0], trial_id)
+        
+        # Final cleanup
+        torch.cuda.empty_cache()
+        progress(1.0, desc="Complete!")
+        
+        return state, video_path, glb_path, glb_path
+        
+    except Exception as e:
+        # Clean up on error
+        torch.cuda.empty_cache()
+        raise gr.Error(f"Processing failed: {str(e)}")
 
-    # Event handlers
-    demo.load(start_session)
-    demo.unload(end_session)
+@spaces.GPU
+def extract_reduced_glb(
+    state: dict,
+    mesh_simplify: float,
+    texture_size: int,
+    req: gr.Request,
+    progress: gr.Progress = gr.Progress()
+) -> Tuple[str, str]:
+    """
+    Extract a reduced-quality GLB file with progress tracking.
+    """
+    user_dir = os.path.join(TMP_DIR, str(req.session_hash))
     
-    image_prompt.upload(
-        preprocess_image,
-        inputs=[image_prompt],
-        outputs=[image_prompt],
-    )
+    try:
+        progress(0.1, desc="Unpacking model state...")
+        gs, mesh, trial_id = unpack_state(state)
+        
+        progress(0.3, desc="Generating reduced GLB...")
+        glb = postprocessing_utils.to_glb(
+            gs, mesh,
+            simplify=mesh_simplify,
+            texture_size=texture_size,
+            verbose=False
+        )
+        
+        progress(0.8, desc="Saving reduced GLB...")
+        glb_path = os.path.join(user_dir, f"{trial_id}_reduced.glb")
+        glb.export(glb_path)
+        
+        progress(0.9, desc="Cleaning up...")
+        torch.cuda.empty_cache()
+        
+        progress(1.0, desc="Complete!")
+        return glb_path, glb_path
+        
+    except Exception as e:
+        torch.cuda.empty_cache()
+        raise gr.Error(f"GLB reduction failed: {str(e)}")
 
-    generate_btn.click(
-        get_seed,
-        inputs=[randomize_seed, seed],
-        outputs=[seed],
-    ).then(
-        image_to_3d,
-        inputs=[image_prompt, seed, ss_guidance_strength, ss_sampling_steps, slat_guidance_strength, slat_sampling_steps],
-        outputs=[video_output, model_output, download_glb],
-    ).then(
-        lambda: gr.Button(interactive=True),
-        outputs=[download_glb],
-    )
+# ... (rest of the UI code remains the same) ...
 
-# Launch the Gradio app
+# Add some memory optimization settings at startup
 if __name__ == "__main__":
+    # Set some CUDA memory management options
+    torch.cuda.empty_cache()
+    torch.backends.cudnn.benchmark = True
+    
+    # Initialize pipeline
     pipeline = TrellisImageTo3DPipeline.from_pretrained("JeffreyXiang/TRELLIS-image-large")
     pipeline.cuda()
+    
     try:
-        pipeline.preprocess_image(Image.fromarray(np.zeros((512, 512, 3), dtype=np.uint8)))    # Preload rembg
+        # Preload rembg with minimal memory usage
+        test_img = np.zeros((256, 256, 3), dtype=np.uint8)  # Smaller test image
+        pipeline.preprocess_image(Image.fromarray(test_img))
+        del test_img
+        torch.cuda.empty_cache()
     except:
         pass
+    
     demo.launch()