update

app.py

@@ -49,17 +49,32 @@ EXAMPLE_APOSE_IMAGES = glob.glob("./input_cases_apose/*")
 
 infer_api = InferAPI(config_canocalize, config_multiview, config_slrm, config_refine)
 
-REMINDER = """
-### Reminder:
-1. **Reference Image**:  
-   - You can upload any reference image (with or without background).  
-   - If the image has an alpha channel (transparency), background segmentation will be automatically performed.  
-   - Alternatively, you can pre-segment the background using other tools and upload the result directly.  
-   - A-pose images are also supported.
-
+_HEADER_ = '''
+<h2><b>[CVPR 2025] StdGEN 🤗 Gradio Demo</b></h2>
+This is official demo for our CVPR 2025 paper <a href="">StdGEN: Semantic-Decomposed 3D Character Generation from Single Images</a>.
+Code: <a href='https://github.com/hyz317/StdGEN' target='_blank'>GitHub</a>. Paper: <a href='https://arxiv.org/abs/2411.05738' target='_blank'>ArXiv</a>.
+❗️❗️❗️**Important Notes:**
+1. - You can upload any reference image (with or without background). A-pose images are also supported (white bkg required).
+   - If the image has an alpha channel (transparency), background segmentation will be automatically performed. Alternatively, you can pre-segment the background using other tools and upload the result directly.  
 2. Real person images generally work well, but note that normals may appear smoother than expected. You can try to use other monocular normal estimation models.
-
 3. The base human model in the output is uncolored due to potential NSFW concerns. If you need colored results, please refer to the official GitHub repository for instructions.
+'''
+
+_CITE_ = r"""
+If StdGEN is helpful, please help to ⭐ the <a href='https://github.com/hyz317/StdGEN' target='_blank'>Github Repo</a>. Thanks! [![GitHub Stars](https://img.shields.io/github/stars/hyz317/StdGEN?style=social)](https://github.com/hyz317/StdGEN)
+---
+📝 **Citation**
+If you find our work useful for your research or applications, please cite using this bibtex:
+```bibtex
+@article{he2024stdgen,
+  title={StdGEN: Semantic-Decomposed 3D Character Generation from Single Images},
+  author={He, Yuze and Zhou, Yanning and Zhao, Wang and Wu, Zhongkai and Xiao, Kaiwen and Yang, Wei and Liu, Yong-Jin and Han, Xiao},
+  journal={arXiv preprint arXiv:2411.05738},
+  year={2024}
+}
+```
+📧 **Contact**
+If you have any questions, feel free to open a discussion or contact us at <b>hyz22@mails.tsinghua.edu.cn</b>.
 """
 
 # 示例占位函数 - 需替换实际模型
@@ -86,7 +101,7 @@ def refine_mesh(apose_img, mesh1, mesh2, mesh3, seed):
     return refined
 
 with gr.Blocks(title="StdGEN: Semantically Decomposed 3D Character Generation from Single Images") as demo:
-    gr.Markdown(REMINDER)
+    gr.Markdown(_HEADER_)
     with gr.Row():
         with gr.Column():
             gr.Markdown("## 1. Reference Image to A-pose Image")
@@ -139,6 +154,8 @@ with gr.Blocks(title="StdGEN: Semantically Decomposed 3D Character Generation fr
         refined_meshes = [gr.Model3D(label=f"refined mesh {i+1}", height=384) for i in range(3)]
         refined_full_mesh = gr.Model3D(label="refined whole mesh", height=384)
 
+    gr.Markdown(_CITE_)
+
     # 交互逻辑
     pose_btn.click(
         arbitrary_to_apose,

refine/render.py

@@ -33,7 +33,7 @@ class NormalsRenderer:
         else:
             self._mvp = mvp
         self._image_size = image_size
-        self._glctx = glctx
+        self._glctx = dr.RasterizeCudaContext(device=device)
         _warmup(self._glctx, device)
 
     def render(self,