restyle launch codes

app.py

@@ -1,319 +1,16 @@
-import os
-import shutil
-import tempfile
-
 import gradio as gr
-import numpy as np
-import rembg
-import spaces
-import torch
-from diffusers import DiffusionPipeline, EulerAncestralDiscreteScheduler, StableDiffusionXLPipeline, EulerDiscreteScheduler
-from einops import rearrange
-from huggingface_hub import hf_hub_download
-from omegaconf import OmegaConf
-from PIL import Image
-from pytorch_lightning import seed_everything
-from torchvision.transforms import v2
-from safetensors.torch import load_file
-
-from src.utils.camera_util import (FOV_to_intrinsics, get_circular_camera_poses,
-                                   get_zero123plus_input_cameras)
-from src.utils.infer_util import (remove_background, resize_foreground)
-from src.utils.mesh_util import save_glb, save_obj
-from src.utils.train_util import instantiate_from_config
-
-
-def find_cuda():
-    cuda_home = os.environ.get('CUDA_HOME') or os.environ.get('CUDA_PATH')
-    if cuda_home and os.path.exists(cuda_home):
-        return cuda_home
-
-    nvcc_path = shutil.which('nvcc')
-    if nvcc_path:
-        cuda_path = os.path.dirname(os.path.dirname(nvcc_path))
-        return cuda_path
-
-    return None
-
-
-def get_render_cameras(batch_size=1, M=120, radius=2.5, elevation=10.0, is_flexicubes=False):
-    c2ws = get_circular_camera_poses(M=M, radius=radius, elevation=elevation)
-    if is_flexicubes:
-        cameras = torch.linalg.inv(c2ws)
-        cameras = cameras.unsqueeze(0).repeat(batch_size, 1, 1, 1)
-    else:
-        extrinsics = c2ws.flatten(-2)
-        intrinsics = FOV_to_intrinsics(50.0).unsqueeze(
-            0).repeat(M, 1, 1).float().flatten(-2)
-        cameras = torch.cat([extrinsics, intrinsics], dim=-1)
-        cameras = cameras.unsqueeze(0).repeat(batch_size, 1, 1)
-    return cameras
-
-
-def check_input_image(input_image):
-    if input_image is None:
-        raise gr.Error("No image selected!")
-
-
-def preprocess(input_image):
-    rembg_session = rembg.new_session()
-
-    input_image = remove_background(input_image, rembg_session)
-    input_image = resize_foreground(input_image, 0.85)
-
-    return input_image
-
-
-def generate_prompt(subject, style, color_scheme, angle, lighting_type, additional_details):
-    prompt = f"A 3D cartoon render of {subject}, featuring the entire body and shape, on a transparent background. The style should be {style}, with {color_scheme} colors, emphasizing the essential features and lines. The pose should clearly showcase the full form of the {subject} from a {angle} perspective. Lighting is {lighting_type}, highlighting the volume and depth of the subject. {additional_details}. Output as a high-resolution PNG with no background."
-    return prompt
-
-
-@spaces.GPU
-def generate_image(subject, style, color_scheme, angle, lighting_type, additional_details):
-    checkpoint = "sdxl_lightning_8step_unet.safetensors"
-    num_inference_steps = 8
-
-    pipe.scheduler = EulerDiscreteScheduler.from_config(
-        pipe.scheduler.config, timestep_spacing="trailing")
-    pipe.unet.load_state_dict(
-        load_file(hf_hub_download(repo, checkpoint), device="cuda"))
-
-    prompt = generate_prompt(subject, style, color_scheme,
-                             angle, lighting_type, additional_details)
-    results = pipe(
-        prompt, num_inference_steps=num_inference_steps, guidance_scale=0)
-    return results.images[0]
-
-
-@spaces.GPU
-def generate_mvs(input_image, sample_steps, sample_seed):
-    seed_everything(sample_seed)
-
-    z123_image = pipeline(
-        input_image, num_inference_steps=sample_steps).images[0]
-
-    show_image = np.asarray(z123_image, dtype=np.uint8)
-    show_image = torch.from_numpy(show_image)
-    show_image = rearrange(
-        show_image, '(n h) (m w) c -> (n m) h w c', n=3, m=2)
-    show_image = rearrange(
-        show_image, '(n m) h w c -> (n h) (m w) c', n=2, m=3)
-    show_image = Image.fromarray(show_image.numpy())
 
-    return z123_image, show_image
+from launch.image_generation import image_generation_ui
+from launch.model_generation import model_generation_ui
 
 
-@spaces.GPU
-def make3d(images):
-    global model
-    if IS_FLEXICUBES:
-        model.init_flexicubes_geometry(device, use_renderer=False)
-    model = model.eval()
-
-    images = np.asarray(images, dtype=np.float32) / 255.0
-    images = torch.from_numpy(images).permute(2, 0, 1).contiguous().float()
-    images = rearrange(images, 'c (n h) (m w) -> (n m) c h w', n=3, m=2)
-
-    input_cameras = get_zero123plus_input_cameras(
-        batch_size=1, radius=4.0).to(device)
-    render_cameras = get_render_cameras(
-        batch_size=1, radius=2.5, is_flexicubes=IS_FLEXICUBES).to(device)
-
-    images = images.unsqueeze(0).to(device)
-    images = v2.functional.resize(
-        images, (320, 320), interpolation=3, antialias=True).clamp(0, 1)
-
-    mesh_fpath = tempfile.NamedTemporaryFile(suffix=f".obj", delete=False).name
-    print(mesh_fpath)
-    mesh_basename = os.path.basename(mesh_fpath).split('.')[0]
-    mesh_dirname = os.path.dirname(mesh_fpath)
-    mesh_glb_fpath = os.path.join(mesh_dirname, f"{mesh_basename}.glb")
-
-    with torch.no_grad():
-        planes = model.forward_planes(images, input_cameras)
-        mesh_out = model.extract_mesh(
-            planes, use_texture_map=False, **infer_config)
-
-        vertices, faces, vertex_colors = mesh_out
-        vertices = vertices[:, [1, 2, 0]]
-
-        save_glb(vertices, faces, vertex_colors, mesh_glb_fpath)
-        save_obj(vertices, faces, vertex_colors, mesh_fpath)
-
-        print(f"Mesh saved to {mesh_fpath}")
-
-    return mesh_fpath, mesh_glb_fpath
-
-
-# Configuration
-cuda_path = find_cuda()
-config_path = 'configs/instant-mesh-large.yaml'
-config = OmegaConf.load(config_path)
-config_name = os.path.basename(config_path).replace('.yaml', '')
-model_config = config.model_config
-infer_config = config.infer_config
-
-IS_FLEXICUBES = config_name.startswith('instant-mesh')
-device = torch.device('cuda')
-
-# Load diffusion model
-print('Loading diffusion model ...')
-pipeline = DiffusionPipeline.from_pretrained(
-    "sudo-ai/zero123plus-v1.2",
-    custom_pipeline="zero123plus",
-    torch_dtype=torch.float16,
-)
-pipeline.scheduler = EulerAncestralDiscreteScheduler.from_config(
-    pipeline.scheduler.config, timestep_spacing='trailing'
-)
-
-unet_ckpt_path = hf_hub_download(
-    repo_id="TencentARC/InstantMesh", filename="diffusion_pytorch_model.bin", repo_type="model")
-state_dict = torch.load(unet_ckpt_path, map_location='cpu')
-pipeline.unet.load_state_dict(state_dict, strict=True)
-
-pipeline = pipeline.to(device)
-
-# Load reconstruction model
-print('Loading reconstruction model ...')
-model_ckpt_path = hf_hub_download(
-    repo_id="TencentARC/InstantMesh", filename="instant_mesh_large.ckpt", repo_type="model")
-model = instantiate_from_config(model_config)
-state_dict = torch.load(model_ckpt_path, map_location='cpu')['state_dict']
-state_dict = {k[14:]: v for k, v in state_dict.items() if k.startswith(
-    'lrm_generator.') and 'source_camera' not in k}
-model.load_state_dict(state_dict, strict=True)
-
-model = model.to(device)
-
-# Load StableDiffusionXL model
-base = "stabilityai/stable-diffusion-xl-base-1.0"
-repo = "ByteDance/SDXL-Lightning"
-
-pipe = StableDiffusionXLPipeline.from_pretrained(
-    base, torch_dtype=torch.float16, variant="fp16").to("cuda")
-
-print('Loading Finished!')
-
 with gr.Blocks() as demo:
     with gr.Group():
         with gr.Tab("Generate Image and Remove Background"):
-            with gr.Row():
-                subject = gr.Textbox(label='Subject', scale=2)
-                style = gr.Dropdown(
-                    label='Style',
-                    choices=['Pixar-like', 'Disney-esque', 'Anime-inspired'],
-                    value='Pixar-like',
-                    multiselect=False,
-                    scale=2
-                )
-                color_scheme = gr.Dropdown(
-                    label='Color Scheme',
-                    choices=['Vibrant', 'Pastel',
-                             'Monochromatic', 'Black and White'],
-                    value='Vibrant',
-                    multiselect=False,
-                    scale=2
-                )
-                angle = gr.Dropdown(
-                    label='Angle',
-                    choices=['Front', 'Side', 'Three-quarter'],
-                    value='Front',
-                    multiselect=False,
-                    scale=2
-                )
-                lighting_type = gr.Dropdown(
-                    label='Lighting Type',
-                    choices=['Bright and Even',
-                             'Dramatic Shadows', 'Soft and Warm'],
-                    value='Bright and Even',
-                    multiselect=False,
-                    scale=2
-                )
-                additional_details = gr.Textbox(
-                    label='Additional Details', scale=2)
-                submit_prompt = gr.Button(
-                    'Generate Image', scale=1, variant='primary')
-
-            with gr.Row(variant="panel"):
-                with gr.Column():
-                    with gr.Row():
-                        input_image = gr.Image(
-                            label="Input Image",
-                            image_mode="RGBA",
-                            sources="upload",
-                            type="pil",
-                            elem_id="content_image",
-                        )
-                        processed_image = gr.Image(
-                            label="Processed Image",
-                            image_mode="RGBA",
-                            type="pil",
-                            interactive=False
-                        )
-                    with gr.Row():
-                        submit_process = gr.Button(
-                            "Remove Background", elem_id="process", variant="primary")
-                    with gr.Row(variant="panel"):
-                        gr.Examples(
-                            examples=[os.path.join("examples", img_name) for img_name in sorted(
-                                os.listdir("examples"))],
-                            inputs=[input_image],
-                            label="Examples",
-                            cache_examples=False,
-                            examples_per_page=16
-                        )
+            input_image, processed_image = image_generation_ui()
 
         with gr.Tab("Generate 3D Model"):
-            with gr.Column():
-                with gr.Row():
-                    with gr.Column():
-                        mv_show_images = gr.Image(
-                            label="Generated Multi-views",
-                            type="pil",
-                            width=379,
-                            interactive=False
-                        )
-                with gr.Row():
-                    with gr.Group():
-                        sample_seed = gr.Number(
-                            value=42, label="Seed Value", precision=0)
-                        sample_steps = gr.Slider(
-                            label="Sample Steps", minimum=30, maximum=75, value=75, step=5)
-                with gr.Row():
-                    submit_mesh = gr.Button(
-                        "Generate 3D Model", elem_id="generate", variant="primary")
-                with gr.Row():
-                    with gr.Tab("OBJ"):
-                        output_model_obj = gr.Model3D(
-                            label="Output Model (OBJ Format)",
-                            interactive=False,
-                        )
-                        gr.Markdown(
-                            "Note: Downloaded .obj model will be flipped. Export .glb instead or manually flip it before usage.")
-                    with gr.Tab("GLB"):
-                        output_model_glb = gr.Model3D(
-                            label="Output Model (GLB Format)",
-                            interactive=False,
-                        )
-                        gr.Markdown(
-                            "Note: The model shown here has a darker appearance. Download to get correct results.")
-                with gr.Row():
-                    gr.Markdown(
-                        '''Try a different <b>seed value</b> if the result is unsatisfying (Default: 42).''')
-
-    mv_images = gr.State()
-
-    submit_prompt.click(fn=generate_image, inputs=[subject, style, color_scheme, angle, lighting_type, additional_details], outputs=input_image).success(
-        fn=preprocess, inputs=[input_image], outputs=[processed_image]
-    )
-    submit_process.click(fn=check_input_image, inputs=[input_image]).success(
-        fn=preprocess, inputs=[input_image], outputs=[processed_image],
-    )
-    submit_mesh.click(fn=generate_mvs, inputs=[processed_image, sample_steps, sample_seed], outputs=[mv_images, mv_show_images]).success(
-        fn=make3d, inputs=[mv_images], outputs=[
-            output_model_obj, output_model_glb]
-    )
+            output_model_obj, output_model_glb = model_generation_ui(
+                processed_image)
 
 demo.launch()

launch/image_generation.py

@@ -0,0 +1,127 @@
+import os
+
+import gradio as gr
+import rembg
+import spaces
+import torch
+from diffusers import StableDiffusionXLPipeline, EulerDiscreteScheduler
+from huggingface_hub import hf_hub_download
+from safetensors.torch import load_file
+
+from src.utils.infer_util import (remove_background, resize_foreground)
+
+
+# Load StableDiffusionXL model
+base = "stabilityai/stable-diffusion-xl-base-1.0"
+repo = "ByteDance/SDXL-Lightning"
+
+pipe = StableDiffusionXLPipeline.from_pretrained(
+    base, torch_dtype=torch.float16, variant="fp16").to("cuda")
+
+
+def generate_prompt(subject, style, color_scheme, angle, lighting_type, additional_details):
+    return f"A 3D cartoon render of {subject}, featuring the entire body and shape, on a transparent background. The style should be {style}, with {color_scheme} colors, emphasizing the essential features and lines. The pose should clearly showcase the full form of the {subject} from a {angle} perspective. Lighting is {lighting_type}, highlighting the volume and depth of the subject. {additional_details}. Output as a high-resolution PNG with no background."
+
+
+@spaces.GPU
+def generate_image(subject, style, color_scheme, angle, lighting_type, additional_details):
+    checkpoint = "sdxl_lightning_8step_unet.safetensors"
+    num_inference_steps = 8
+
+    pipe.scheduler = EulerDiscreteScheduler.from_config(
+        pipe.scheduler.config, timestep_spacing="trailing")
+    pipe.unet.load_state_dict(
+        load_file(hf_hub_download(repo, checkpoint), device="cuda"))
+
+    prompt = generate_prompt(subject, style, color_scheme,
+                             angle, lighting_type, additional_details)
+    results = pipe(
+        prompt, num_inference_steps=num_inference_steps, guidance_scale=0)
+    return results.images[0]
+
+
+def check_input_image(input_image):
+    if input_image is None:
+        raise gr.Error("No image selected!")
+
+
+def preprocess(input_image):
+    rembg_session = rembg.new_session()
+
+    input_image = remove_background(input_image, rembg_session)
+    input_image = resize_foreground(input_image, 0.85)
+
+    return input_image
+
+
+def image_generation_ui():
+    with gr.Row():
+        subject = gr.Textbox(label='Subject', scale=2)
+        style = gr.Dropdown(
+            label='Style',
+            choices=['Pixar-like', 'Disney-esque', 'Anime-inspired'],
+            value='Pixar-like',
+            multiselect=False,
+            scale=2
+        )
+        color_scheme = gr.Dropdown(
+            label='Color Scheme',
+            choices=['Vibrant', 'Pastel', 'Monochromatic', 'Black and White'],
+            value='Vibrant',
+            multiselect=False,
+            scale=2
+        )
+        angle = gr.Dropdown(
+            label='Angle',
+            choices=['Front', 'Side', 'Three-quarter'],
+            value='Front',
+            multiselect=False,
+            scale=2
+        )
+        lighting_type = gr.Dropdown(
+            label='Lighting Type',
+            choices=['Bright and Even', 'Dramatic Shadows', 'Soft and Warm'],
+            value='Bright and Even',
+            multiselect=False,
+            scale=2
+        )
+        additional_details = gr.Textbox(label='Additional Details', scale=2)
+        submit_prompt = gr.Button('Generate Image', scale=1, variant='primary')
+
+    with gr.Row(variant="panel"):
+        with gr.Column():
+            with gr.Row():
+                input_image = gr.Image(
+                    label="Input Image",
+                    image_mode="RGBA",
+                    sources="upload",
+                    type="pil",
+                    elem_id="content_image",
+                )
+                processed_image = gr.Image(
+                    label="Processed Image",
+                    image_mode="RGBA",
+                    type="pil",
+                    interactive=False
+                )
+            with gr.Row():
+                submit_process = gr.Button(
+                    "Remove Background", elem_id="process", variant="primary")
+            with gr.Row(variant="panel"):
+                gr.Examples(
+                    examples=[os.path.join("examples", img_name)
+                              for img_name in sorted(os.listdir("examples"))],
+                    inputs=[input_image],
+                    label="Examples",
+                    cache_examples=False,
+                    examples_per_page=16
+                )
+
+    submit_prompt.click(fn=generate_image, inputs=[subject, style, color_scheme, angle, lighting_type, additional_details], outputs=input_image).success(
+        fn=preprocess, inputs=[input_image], outputs=[processed_image]
+    )
+    submit_process.click(fn=check_input_image, inputs=[input_image]).success(
+        fn=preprocess, inputs=[input_image], outputs=[processed_image],
+    )
+
+    return input_image, processed_image

launch/model_generation.py

@@ -0,0 +1,183 @@
+import os
+import tempfile
+
+import gradio as gr
+import numpy as np
+from launch.utils import find_cuda
+import spaces
+import torch
+from diffusers import DiffusionPipeline, EulerAncestralDiscreteScheduler
+from einops import rearrange
+from huggingface_hub import hf_hub_download
+from omegaconf import OmegaConf
+from PIL import Image
+from pytorch_lightning import seed_everything
+from torchvision.transforms import v2
+
+from src.utils.camera_util import (FOV_to_intrinsics, get_circular_camera_poses,
+                                   get_zero123plus_input_cameras)
+from src.utils.mesh_util import save_glb, save_obj
+from src.utils.train_util import instantiate_from_config
+
+# Configuration
+cuda_path = find_cuda()
+config_path = 'configs/instant-mesh-large.yaml'
+config = OmegaConf.load(config_path)
+config_name = os.path.basename(config_path).replace('.yaml', '')
+model_config = config.model_config
+infer_config = config.infer_config
+
+IS_FLEXICUBES = config_name.startswith('instant-mesh')
+device = torch.device('cuda')
+
+# Load diffusion model
+print('Loading diffusion model ...')
+pipeline = DiffusionPipeline.from_pretrained(
+    "sudo-ai/zero123plus-v1.2",
+    custom_pipeline="zero123plus",
+    torch_dtype=torch.float16,
+)
+pipeline.scheduler = EulerAncestralDiscreteScheduler.from_config(
+    pipeline.scheduler.config, timestep_spacing='trailing'
+)
+
+unet_ckpt_path = hf_hub_download(
+    repo_id="TencentARC/InstantMesh", filename="diffusion_pytorch_model.bin", repo_type="model")
+state_dict = torch.load(unet_ckpt_path, map_location='cpu')
+pipeline.unet.load_state_dict(state_dict, strict=True)
+
+pipeline = pipeline.to(device)
+
+# Load reconstruction model
+print('Loading reconstruction model ...')
+model_ckpt_path = hf_hub_download(
+    repo_id="TencentARC/InstantMesh", filename="instant_mesh_large.ckpt", repo_type="model")
+model = instantiate_from_config(model_config)
+state_dict = torch.load(model_ckpt_path, map_location='cpu')['state_dict']
+state_dict = {k[14:]: v for k, v in state_dict.items() if k.startswith(
+    'lrm_generator.') and 'source_camera' not in k}
+model.load_state_dict(state_dict, strict=True)
+
+model = model.to(device)
+
+
+def get_render_cameras(batch_size=1, M=120, radius=2.5, elevation=10.0, is_flexicubes=False):
+    c2ws = get_circular_camera_poses(M=M, radius=radius, elevation=elevation)
+    if is_flexicubes:
+        cameras = torch.linalg.inv(c2ws)
+        cameras = cameras.unsqueeze(0).repeat(batch_size, 1, 1, 1)
+    else:
+        extrinsics = c2ws.flatten(-2)
+        intrinsics = FOV_to_intrinsics(50.0).unsqueeze(
+            0).repeat(M, 1, 1).float().flatten(-2)
+        cameras = torch.cat([extrinsics, intrinsics], dim=-1)
+        cameras = cameras.unsqueeze(0).repeat(batch_size, 1, 1)
+    return cameras
+
+
+@spaces.GPU
+def generate_mvs(input_image, sample_steps, sample_seed):
+    seed_everything(sample_seed)
+
+    z123_image = pipeline(
+        input_image, num_inference_steps=sample_steps).images[0]
+
+    show_image = np.asarray(z123_image, dtype=np.uint8)
+    show_image = torch.from_numpy(show_image)
+    show_image = rearrange(
+        show_image, '(n h) (m w) c -> (n m) h w c', n=3, m=2)
+    show_image = rearrange(
+        show_image, '(n m) h w c -> (n h) (m w) c', n=2, m=3)
+    show_image = Image.fromarray(show_image.numpy())
+
+    return z123_image, show_image
+
+
+@spaces.GPU
+def make3d(images):
+    global model
+    if IS_FLEXICUBES:
+        model.init_flexicubes_geometry(device, use_renderer=False)
+    model = model.eval()
+
+    images = np.asarray(images, dtype=np.float32) / 255.0
+    images = torch.from_numpy(images).permute(2, 0, 1).contiguous().float()
+    images = rearrange(images, 'c (n h) (m w) -> (n m) c h w', n=3, m=2)
+
+    input_cameras = get_zero123plus_input_cameras(
+        batch_size=1, radius=4.0).to(device)
+    render_cameras = get_render_cameras(
+        batch_size=1, radius=2.5, is_flexicubes=IS_FLEXICUBES).to(device)
+
+    images = images.unsqueeze(0).to(device)
+    images = v2.functional.resize(
+        images, (320, 320), interpolation=3, antialias=True).clamp(0, 1)
+
+    mesh_fpath = tempfile.NamedTemporaryFile(suffix=f".obj", delete=False).name
+    print(mesh_fpath)
+    mesh_basename = os.path.basename(mesh_fpath).split('.')[0]
+    mesh_dirname = os.path.dirname(mesh_fpath)
+    mesh_glb_fpath = os.path.join(mesh_dirname, f"{mesh_basename}.glb")
+
+    with torch.no_grad():
+        planes = model.forward_planes(images, input_cameras)
+        mesh_out = model.extract_mesh(
+            planes, use_texture_map=False, **infer_config)
+
+        vertices, faces, vertex_colors = mesh_out
+        vertices = vertices[:, [1, 2, 0]]
+
+        save_glb(vertices, faces, vertex_colors, mesh_glb_fpath)
+        save_obj(vertices, faces, vertex_colors, mesh_fpath)
+
+        print(f"Mesh saved to {mesh_fpath}")
+
+    return mesh_fpath, mesh_glb_fpath
+
+
+def model_generation_ui(processed_image):
+    with gr.Column():
+        with gr.Row():
+            with gr.Column():
+                mv_show_images = gr.Image(
+                    label="Generated Multi-views",
+                    type="pil",
+                    width=379,
+                    interactive=False
+                )
+        with gr.Row():
+            with gr.Group():
+                sample_seed = gr.Number(
+                    value=42, label="Seed Value", precision=0)
+                sample_steps = gr.Slider(
+                    label="Sample Steps", minimum=30, maximum=75, value=75, step=5)
+        with gr.Row():
+            submit_mesh = gr.Button(
+                "Generate 3D Model", elem_id="generate", variant="primary")
+        with gr.Row():
+            with gr.Tab("OBJ"):
+                output_model_obj = gr.Model3D(
+                    label="Output Model (OBJ Format)",
+                    interactive=False,
+                )
+                gr.Markdown(
+                    "Note: Downloaded .obj model will be flipped. Export .glb instead or manually flip it before usage.")
+            with gr.Tab("GLB"):
+                output_model_glb = gr.Model3D(
+                    label="Output Model (GLB Format)",
+                    interactive=False,
+                )
+                gr.Markdown(
+                    "Note: The model shown here has a darker appearance. Download to get correct results.")
+        with gr.Row():
+            gr.Markdown(
+                '''Try a different <b>seed value</b> if the result is unsatisfying (Default: 42).''')
+
+    mv_images = gr.State()
+
+    submit_mesh.click(fn=generate_mvs, inputs=[processed_image, sample_steps, sample_seed], outputs=[mv_images, mv_show_images]).success(
+        fn=make3d, inputs=[mv_images], outputs=[
+            output_model_obj, output_model_glb]
+    )
+
+    return output_model_obj, output_model_glb

launch/utils.py

@@ -0,0 +1,14 @@
+import os
+import shutil
+
+def find_cuda():
+    cuda_home = os.environ.get('CUDA_HOME') or os.environ.get('CUDA_PATH')
+    if cuda_home and os.path.exists(cuda_home):
+        return cuda_home
+
+    nvcc_path = shutil.which('nvcc')
+    if nvcc_path:
+        cuda_path = os.path.dirname(os.path.dirname(nvcc_path))
+        return cuda_path
+
+    return None