run demo.py via file path, not -m

app.py

@@ -1,195 +1,44 @@
-import subprocess, uuid, os, sys
+import subprocess, uuid, os
 from pathlib import Path
 import gradio as gr
 
 # ── paths ─────────────────────────────────────────────────────
-CKPT   = "checkpoints/mld_humanml.pt"         # your 90 MB weight
-DEVICE = "cpu"                                # free HF Spaces
-STEPS  = "50"                                 # diffusion steps
-CFG    = "motion_latent_diffusion/configs/config_mld_humanml3d.yaml"  # default config file
+CKPT   = "checkpoints/mld_humanml.pt"       # your 90 MB weight
+DEVICE = "cpu"                              # free HF Spaces
+STEPS  = "50"                               # diffusion steps
+CFG    = "configs/config_mld_humanml3d.yaml"  # default config file
 # ──────────────────────────────────────────────────────────────
 
-def ensure_dependencies():
-    """Ensure all required packages are installed"""
-    required_packages = [
-        "omegaconf", 
-        "torch",
-        "numpy", 
-        "tqdm",
-        "einops"
-    ]
-    for package in required_packages:
-        try:
-            __import__(package)
-        except ImportError:
-            print(f"Installing {package}...")
-            subprocess.check_call([sys.executable, "-m", "pip", "install", package])
-
 def generate_motion(prompt: str) -> str:
-    """Generate motion from text prompt and return path to BVH file"""
-    ensure_dependencies()
-    
-    # Create a unique temporary file for output
     out_bvh = Path("/tmp") / f"{uuid.uuid4().hex}.bvh"
-    
-    # Get absolute paths
-    root = Path(__file__).parent.absolute()
-    demo_script = root / "motion_latent_diffusion" / "demo.py"
-    config_path = root / CFG
-    ckpt_path = root / CKPT
-    
-    # Log paths for debugging
-    print(f"Root directory: {root}")
-    print(f"Demo script: {demo_script}")
-    print(f"Config file: {config_path}")
-    print(f"Checkpoint: {ckpt_path}")
-    
-    # Check that files exist
-    if not demo_script.exists():
-        return f"Error: Demo script not found at {demo_script}"
-    if not config_path.exists():
-        return f"Error: Config file not found at {config_path}"
-    if not ckpt_path.exists():
-        return f"Error: Checkpoint not found at {ckpt_path}"
-    
-    # Run the demo script directly
+
     cmd = [
-        sys.executable,  # Use the current Python interpreter
-        str(demo_script),
-        "--cfg", str(config_path),
-        "--checkpoint", str(ckpt_path),
-        "--prompt", prompt,
-        "--device", DEVICE,
-        "--steps", STEPS,
-        "--output", str(out_bvh)
+        "python",
+        "-m", "motion_latent_diffusion/demo.py",   # ← new: run demo.py
+        "--cfg",        CFG,
+        "--checkpoint", CKPT,
+        "--prompt",     prompt,
+        "--device",     DEVICE,
+        "--steps",      STEPS,
+        "--output",     str(out_bvh)
     ]
 
-    # Set up environment
+    # make repo importable
     env = os.environ.copy()
-    
-    # Add the repository root to PYTHONPATH so imports work
-    mld_path = root / "motion_latent_diffusion"
-    env["PYTHONPATH"] = f"{env.get('PYTHONPATH', '')}:{root}:{mld_path}"
-    
-    print(f"Running command: {' '.join(cmd)}")
-    print(f"PYTHONPATH: {env['PYTHONPATH']}")
-    
-    try:
-        # Run the command and capture output
-        result = subprocess.run(
-            cmd, 
-            env=env, 
-            check=True, 
-            stderr=subprocess.PIPE, 
-            stdout=subprocess.PIPE,
-            text=True
-        )
-        
-        print("Command output:", result.stdout)
-        
-        # Check if the BVH file was created
-        if out_bvh.exists():
-            print(f"BVH file created at {out_bvh}")
-            return str(out_bvh)
-        else:
-            return f"Error: BVH file not created. Output: {result.stdout}\nError: {result.stderr}"
-            
-    except subprocess.CalledProcessError as e:
-        error_message = f"Command failed with exit code {e.returncode}.\nStdout: {e.stdout}\nStderr: {e.stderr}"
-        print(error_message)
-        return f"Error generating motion: {error_message}"
+    root = Path(__file__).parent
+    env["PYTHONPATH"] = f"{env.get('PYTHONPATH','')}:{root}"
 
-# Alternative direct approach if subprocess method fails
-def generate_motion_direct(prompt: str) -> str:
-    """Generate motion using direct API calls instead of subprocess"""
-    ensure_dependencies()
-    
-    out_bvh = Path("/tmp") / f"{uuid.uuid4().hex}.bvh"
-    
-    # Add the repository root and MLD to path
-    root = Path(__file__).parent.absolute()
-    mld_path = root / "motion_latent_diffusion"
-    
-    if str(root) not in sys.path:
-        sys.path.insert(0, str(root))
-    if str(mld_path) not in sys.path:
-        sys.path.insert(0, str(mld_path))
-    
-    try:
-        # Import necessary modules
-        from omegaconf import OmegaConf
-        import torch
-        from motion_latent_diffusion.mld.config import get_cfg_defaults
-        
-        # Load config
-        cfg = get_cfg_defaults()
-        config_path = root / CFG
-        print(f"Loading config from {config_path}")
-        cfg.merge_from_file(str(config_path))
-        
-        # Override config values
-        cfg.TEST.CHECKPOINT = str(root / CKPT)
-        cfg.DEVICE = DEVICE
-        
-        # Import model
-        from motion_latent_diffusion.mld.models import get_model
-        
-        # Load model
-        print("Loading model...")
-        model = get_model(cfg)
-        state_dict = torch.load(cfg.TEST.CHECKPOINT, map_location='cpu')
-        model.load_state_dict(state_dict)
-        model.eval()
-        
-        # Generate motion
-        print(f"Generating motion for prompt: {prompt}")
-        with torch.no_grad():
-            # Tokenize text
-            tokens = model.tokenizer.encode([prompt])
-            tokens = torch.LongTensor(tokens).to(model.device)
-            
-            # Generate motion
-            motion_data = model.sample(tokens, steps=int(STEPS))
-            
-            # Save as BVH
-            from motion_latent_diffusion.mld.data.humanml.scripts.motion_process import recover_from_ric
-            from motion_latent_diffusion.mld.data.humanml.utils.visualization import plot_3d_motion
-            
-            motion = motion_data[0][0]  # First sample
-            motion_np = motion.cpu().numpy()
-            motion_joints = recover_from_ric(motion_np, 22)
-            
-            # Save motion
-            plot_3d_motion(motion_joints, str(out_bvh), save_bvh=True, title=f"Motion from: {prompt}")
-        
-        if out_bvh.exists():
-            print(f"BVH file created at {out_bvh}")
-            return str(out_bvh)
-        else:
-            return f"Error: BVH file not created."
-    
-    except Exception as e:
-        import traceback
-        error_details = traceback.format_exc()
-        print(f"Error in direct generation: {str(e)}\n{error_details}")
-        return f"Error generating motion: {str(e)}"
+    subprocess.run(cmd, env=env, check=True)
+    return str(out_bvh)
 
 # ── Gradio UI ─────────────────────────────────────────────────
 iface = gr.Interface(
-    fn=generate_motion,  # Try the subprocess version first
-    # If subprocess version doesn't work, switch to direct:
-    # fn=generate_motion_direct,
+    fn=generate_motion,
     inputs=gr.Textbox(lines=2, placeholder="e.g. a person walks and waves"),
     outputs=gr.File(label="Download BVH"),
     title="Motion-Latent-Diffusion – Text → BVH (CPU demo)",
-    description="Enter a prompt to generate a 3-second motion clip.",
-    examples=[
-        "a person walks forward and waves",
-        "a person dances happily",
-        "a person jumps up and down",
-        "a person does a backflip"
-    ]
+    description="Enter a prompt to generate a 3-second motion clip."
 )
 
 if __name__ == "__main__":
-    iface.launch()
+    iface.launch()