Update app.py

app.py

@@ -3,73 +3,29 @@ 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
 from easydict import EasyDict as edict
-from PIL import Image, ImageOps
+from PIL import Image
 from trellis.pipelines import TrellisImageTo3DPipeline
 from trellis.representations import Gaussian, MeshExtractResult
 from trellis.utils import render_utils, postprocessing_utils
-from diffusers import ControlNetModel, StableDiffusionXLControlNetPipeline, AutoencoderKL
-from diffusers import EulerAncestralDiscreteScheduler
 
 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)
 
-# Configuración de estilos
-style_list = [
-    {
-        "name": "3D Model",
-        "prompt": "professional 3d model {prompt} . octane render, highly detailed, volumetric, dramatic lighting",
-        "negative_prompt": "ugly, deformed, noisy, low poly, blurry, painting",
-    },
-    # ... (puedes agregar más estilos del ejemplo original si es necesario)
-]
-styles = {k["name"]: (k["prompt"], k["negative_prompt"]) for k in style_list}
-STYLE_NAMES = list(styles.keys())
-DEFAULT_STYLE_NAME = "3D Model"
-
 def start_session(req: gr.Request):
     user_dir = os.path.join(TMP_DIR, str(req.session_hash))
     os.makedirs(user_dir, exist_ok=True)
-    
+
 def end_session(req: gr.Request):
     user_dir = os.path.join(TMP_DIR, str(req.session_hash))
-    shutil.rmtree(user_dir)
+    shutil.rmtree(user_dir, ignore_errors=True)
 
-def apply_style(style_name: str, positive: str, negative: str = "") -> tuple[str, str]:
-    p, n = styles.get(style_name, styles[DEFAULT_STYLE_NAME])
-    return p.replace("{prompt}", positive), n + negative
-
-@spaces.GPU
-def preprocess_image(image: Image.Image, style_name: str, prompt: str) -> Image.Image:
-    # Preprocesamiento con ControlNet
-    width, height = image.size
-    ratio = np.sqrt(1024. * 1024. / (width * height))
-    new_size = (int(width * ratio), int(height * ratio))
-    
-    image = image.resize(new_size)
-    image = ImageOps.invert(image.convert("L")).convert("RGB")
-    
-    prompt, negative_prompt = apply_style(style_name, prompt)
-    
-    # Generación con ControlNet
-    output = pipe_control(
-        prompt=prompt,
-        negative_prompt=negative_prompt,
-        image=image,
-        num_inference_steps=20,
-        controlnet_conditioning_scale=0.85,
-        guidance_scale=5.0,
-        width=new_size[0],
-        height=new_size[1]
-    ).images[0]
-    
-    return output
+def preprocess_image(image: Image.Image) -> Image.Image:
+    return pipeline.preprocess_image(image)
 
 def pack_state(gs: Gaussian, mesh: MeshExtractResult) -> dict:
     return {
@@ -86,195 +42,84 @@ def pack_state(gs: Gaussian, mesh: MeshExtractResult) -> dict:
             'faces': mesh.faces.cpu().numpy(),
         },
     }
-    
+
 def unpack_state(state: dict) -> Tuple[Gaussian, edict]:
-    gs = Gaussian(**state['gaussian']['init_params'])
-    gs._xyz = torch.tensor(state['gaussian']['_xyz']).cuda()
-    gs._features_dc = torch.tensor(state['gaussian']['_features_dc']).cuda()
-    gs._scaling = torch.tensor(state['gaussian']['_scaling']).cuda()
-    gs._rotation = torch.tensor(state['gaussian']['_rotation']).cuda()
-    gs._opacity = torch.tensor(state['gaussian']['_opacity']).cuda()
-    
+    gs = Gaussian(**state['gaussian'])
+    gs._xyz = torch.tensor(state['gaussian']['_xyz'], device='cuda')
+    gs._features_dc = torch.tensor(state['gaussian']['_features_dc'], device='cuda')
+    gs._scaling = torch.tensor(state['gaussian']['_scaling'], device='cuda')
+    gs._rotation = torch.tensor(state['gaussian']['_rotation'], device='cuda')
+    gs._opacity = torch.tensor(state['gaussian']['_opacity'], device='cuda')
+
     mesh = edict(
-        vertices=torch.tensor(state['mesh']['vertices']).cuda(),
-        faces=torch.tensor(state['mesh']['faces']).cuda(),
+        vertices=torch.tensor(state['mesh']['vertices'], device='cuda'),
+        faces=torch.tensor(state['mesh']['faces'], device='cuda'),
     )
-    
     return gs, mesh
 
+def get_seed(randomize_seed: bool, seed: int) -> int:
+    return np.random.randint(0, MAX_SEED) if randomize_seed else seed
+
 @spaces.GPU
-def image_to_3d(
-    image: Image.Image,
-    seed: int,
-    ss_guidance: float,
-    ss_steps: int,
-    slat_guidance: float,
-    slat_steps: int,
-    req: gr.Request,
-) -> Tuple[dict, str]:
+def image_to_3d(image: Image.Image, seed: int, ss_guidance_strength: float, ss_sampling_steps: int, slat_guidance_strength: float, slat_sampling_steps: int, req: gr.Request) -> Tuple[dict, str]:
     user_dir = os.path.join(TMP_DIR, str(req.session_hash))
     outputs = pipeline.run(
         image,
         seed=seed,
         formats=["gaussian", "mesh"],
-        sparse_structure_sampler_params={
-            "steps": ss_steps,
-            "cfg_strength": ss_guidance,
-        },
-        slat_sampler_params={
-            "steps": slat_steps,
-            "cfg_strength": slat_guidance,
-        },
+        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},
     )
-   
+
     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))]
     video_path = os.path.join(user_dir, 'sample.mp4')
     imageio.mimsave(video_path, video, fps=15)
-    
     state = pack_state(outputs['gaussian'][0], outputs['mesh'][0])
     torch.cuda.empty_cache()
     return state, video_path
 
 @spaces.GPU(duration=90)
-def extract_glb(
-    state: dict,
-    simplify: float,
-    texture_size: int,
-    req: gr.Request,
-) -> str:
+def extract_glb(state: dict, mesh_simplify: float, texture_size: int, req: gr.Request) -> Tuple[str, str]:
     user_dir = os.path.join(TMP_DIR, str(req.session_hash))
     gs, mesh = unpack_state(state)
-    
-    glb = postprocessing_utils.to_glb(
-        gs,
-        mesh,
-        simplify=simplify,
-        texture_size=texture_size,
-        verbose=False
-    )
-    
-    glb_path = os.path.join(user_dir, 'model.glb')
+    glb = postprocessing_utils.to_glb(gs, mesh, simplify=mesh_simplify, texture_size=texture_size, verbose=False)
+    glb_path = os.path.join(user_dir, 'sample.glb')
     glb.export(glb_path)
     torch.cuda.empty_cache()
-    return glb_path
+    return glb_path, glb_path
 
 with gr.Blocks() as demo:
-    gr.Markdown("# UTPL - Conversión de Imagen a 3D")
-    gr.Markdown("## Tesis: Modelado 3D automático con IA")
-    
-    with gr.Row():
-        # Columna de entrada
-        with gr.Column(scale=2):
-            image_input = gr.ImageMask(
-                label="Sube tu boceto",
-                type="pil",
-                image_mode="RGB",
-                height=512,
-                value={"background": Image.new("RGB", (512, 512), (255,255,255))}
-            )
-            
-            with gr.Accordion("Configuración de Generación", open=False):
-                style = gr.Dropdown(
-                    STYLE_NAMES,
-                    value=DEFAULT_STYLE_NAME,
-                    label="Estilo"
-                )
-                prompt = gr.Textbox(
-                    label="Prompt",
-                    placeholder="Describe tu modelo 3D"
-                )
-                seed = gr.Slider(0, MAX_SEED, label="Seed", value=0, step=1)
-                randomize_seed = gr.Checkbox(True, label="Semilla Aleatoria")
-                
-                with gr.Group():
-                    gr.Markdown("#### Parámetros 3D")
-                    ss_guidance = gr.Slider(0, 10, 7.5, label="Guía Estructura")
-                    ss_steps = gr.Slider(1, 50, 12, step=1, label="Pasos Estructura")
-                    slat_guidance = gr.Slider(0, 10, 3.0, label="Guía Detalle")
-                    slat_steps = gr.Slider(1, 50, 12, step=1, label="Pasos Detalle")
-            
-            generate_btn = gr.Button("Generar 3D", variant="primary")
-            
-            with gr.Accordion("Exportar GLB", open=False):
-                mesh_simplify = gr.Slider(0.9, 0.98, 0.95, label="Simplificación")
-                texture_size = gr.Slider(512, 2048, 1024, step=512, label="Tamaño Textura")
-                export_glb_btn = gr.Button("Exportar GLB", interactive=False)
-
-        # Columna de salida
-        with gr.Column(scale=3):
-            video_preview = gr.Video(label="Vista Previa 3D", height=400)
-            model_viewer = LitModel3D(label="Visor 3D", height=400)
-            download_glb = gr.DownloadButton("Descargar GLB", interactive=False)
-
-    output_state = gr.State()
-
-    # Carga de pipelines
-    pipe_control = None
-    pipeline = None
+    gr.Markdown("""
+    # Conversión de Imágen a 3D
+    """)
     
-    def initialize_pipelines():
-        global pipe_control, pipeline
-        device = "cuda" if torch.cuda.is_available() else "cpu"
-        
-        # ControlNet setup
-        controlnet = ControlNetModel.from_pretrained(
-            "xinsir/controlnet-scribble-sdxl-1.0",
-            torch_dtype=torch.float16
-        )
-        vae = AutoencoderKL.from_pretrained(
-            "madebyollin/sdxl-vae-fp16-fix",
-            torch_dtype=torch.float16
-        )
-        pipe_control = StableDiffusionXLControlNetPipeline.from_pretrained(
-            "sd-community/sdxl-flash",
-            controlnet=controlnet,
-            vae=vae,
-            torch_dtype=torch.float16
-        ).to(device)
-        pipe_control.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe_control.scheduler.config)
-        
-        # TRELLIS pipeline
-        pipeline = TrellisImageTo3DPipeline.from_pretrained("JeffreyXiang/TRELLIS-image-large").cuda()
+    image_prompt = gr.Image(label="Input Image", type="pil")
+    seed = gr.Slider(0, MAX_SEED, label="Seed", value=0)
+    randomize_seed = gr.Checkbox(label="Randomize Seed", value=True)
+    generate_btn = gr.Button("Generate 3D Asset")
+    video_output = gr.Video(label="3D Preview")
+    model_output = LitModel3D(label="3D Model Viewer")
+    extract_glb_btn = gr.Button("Export GLB")
+    download_glb = gr.DownloadButton(label="Download GLB")
+    output_buf = gr.State()
     
-    demo.load(initialize_pipelines)
-    demo.load(start_session)
-    demo.unload(end_session)
-
-    # Flujo de generación
-    image_processed = gr.Image(visible=False)
-    
-    image_input.upload(
-        preprocess_image,
-        inputs=[image_input, style, prompt],
-        outputs=image_processed
-    )
-
     generate_btn.click(
-        lambda rnd, s: np.random.randint(0, MAX_SEED) if rnd else s,
-        inputs=[randomize_seed, seed],
-        outputs=seed
+        get_seed, inputs=[randomize_seed, seed], outputs=[seed]
     ).then(
-        image_to_3d,
-        inputs=[image_processed, seed, ss_guidance, ss_steps, slat_guidance, slat_steps],
-        outputs=[output_state, video_preview]
+        image_to_3d, inputs=[image_prompt, seed, 7.5, 12, 3.0, 12], outputs=[output_buf, video_output]
     ).then(
-        lambda: gr.update(interactive=True),
-        outputs=export_glb_btn
+        lambda: gr.Button(interactive=True), outputs=[extract_glb_btn]
     )
-
-    export_glb_btn.click(
-        extract_glb,
-        inputs=[output_state, mesh_simplify, texture_size],
-        outputs=model_viewer
+    
+    extract_glb_btn.click(
+        extract_glb, inputs=[output_buf, 0.95, 1024], outputs=[model_output, download_glb]
     ).then(
-        lambda: gr.update(interactive=True),
-        outputs=download_glb
-    )
-
-    model_viewer.change(
-        lambda: gr.update(value=model_viewer.value),
-        outputs=download_glb
+        lambda: gr.Button(interactive=True), outputs=[download_glb]
     )
 
 if __name__ == "__main__":
-    demo.launch()
+    pipeline = TrellisImageTo3DPipeline.from_pretrained("cavargas10/TRELLIS").cuda()
+    demo.launch()