update

app.py

@@ -35,19 +35,13 @@ with gr.Blocks(
                 with gr.Tab(
                     label="Image(auto seg)", id=0
                 ) as single_image_input_tab:
-                    image_prompt_raw = gr.Image(
-                        label="Input Image(raw)",
-                        format="png",
-                        image_mode="RGBA",
-                        type="pil",
-                        height=300,
-                    )
+                    raw_image_cache = gr.State()
                     image_prompt = gr.Image(
                         label="Input Image",
                         format="png",
                         image_mode="RGBA",
                         type="pil",
-                        visible=False,
+                        height=300,
                     )
                     gr.Markdown(
                         """
@@ -59,9 +53,6 @@ with gr.Blocks(
                 ) as samimage_input_tab:
                     with gr.Row():
                         with gr.Column(scale=1):
-                            image_prompt_sam_raw = gr.Image(
-                                label="Input Image(raw)", type="numpy", visible=False,
-                            )
                             image_prompt_sam = gr.Image(
                                 label="Input Image", type="numpy", height=400
                             )
@@ -193,14 +184,14 @@ with gr.Blocks(
                 examples = gr.Examples(
                     label="Image Gallery",
                     examples=[
-                        [f"scripts/apps/assets/example_image/{image}"]
+                        [f"assets/example_image/{image}"]
                         for image in os.listdir(
-                            "scripts/apps/assets/example_image"
+                            "assets/example_image"
                         )
                     ],
-                    inputs=[image_prompt_raw],
+                    inputs=[image_prompt],
                     fn=preprocess_image_fn,
-                    outputs=[image_prompt],
+                    outputs=[image_prompt, raw_image_cache],
                     run_on_click=True,
                     examples_per_page=10,
                 )
@@ -209,14 +200,14 @@ with gr.Blocks(
                 examples = gr.Examples(
                     label="Image Gallery",
                     examples=[
-                        f"scripts/apps/assets/example_image/{image}"
+                        f"assets/example_image/{image}"
                         for image in os.listdir(
-                            "scripts/apps/assets/example_image"
+                            "assets/example_image"
                         )
                     ],
-                    inputs=[image_prompt_sam_raw],
+                    inputs=[image_prompt_sam],
                     fn=preprocess_sam_image_fn,
-                    outputs=[image_prompt_sam],
+                    outputs=[image_prompt_sam, raw_image_cache],
                     run_on_click=True,
                     examples_per_page=10,
                 )
@@ -272,10 +263,10 @@ with gr.Blocks(
         outputs=[is_samimage, single_sam_image_example, single_image_example],
     )
 
-    image_prompt_raw.upload(
+    image_prompt.upload(
         preprocess_image_fn,
-        inputs=[image_prompt_raw],
-        outputs=[image_prompt],
+        inputs=[image_prompt],
+        outputs=[image_prompt, raw_image_cache],
     )
     image_prompt.change(
         lambda: tuple(
@@ -321,12 +312,11 @@ with gr.Blocks(
         outputs=generate_btn,
     )
 
-    image_prompt_sam_raw.upload(
+    image_prompt_sam.upload(
         preprocess_sam_image_fn,
-        inputs=[image_prompt_sam_raw],
-        outputs=[image_prompt_sam],
+        inputs=[image_prompt_sam],
+        outputs=[image_prompt_sam, raw_image_cache],
     )
-
     image_prompt_sam.change(
         lambda: tuple(
             [
@@ -396,6 +386,7 @@ with gr.Blocks(
             ss_sampling_steps,
             slat_guidance_strength,
             slat_sampling_steps,
+            raw_image_cache,
             image_seg_sam,
             is_samimage,
         ],
@@ -448,4 +439,4 @@ with gr.Blocks(
 
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch(server_name="10.34.8.82", server_port=8084)

common.py

@@ -127,20 +127,6 @@ elif os.getenv("GRADIO_APP") == "texture_edit":
 os.makedirs(TMP_DIR, exist_ok=True)
 
 
-def inject_image_css():
-    return gr.HTML(
-        """
-        <style>
-        .image-container img {
-            object-fit: contain !important;
-            max-width: 100% !important;
-            max-height: 100% !important;
-        }
-        </style>
-        """
-    )
-    
-
 def start_session(req: gr.Request) -> None:
     user_dir = os.path.join(TMP_DIR, str(req.session_hash))
     os.makedirs(user_dir, exist_ok=True)
@@ -213,26 +199,27 @@ def preprocess_image_fn(
         image = Image.open(image)
     elif isinstance(image, np.ndarray):
         image = Image.fromarray(image)
+    
+    image_cache = image.copy().resize((512, 512))
 
-    # image.save(f"{TMP_DIR}/{req.session_hash}/raw_image.png")
     image = RBG_REMOVER(image)
     image = trellis_preprocess(image)
 
-    return image
+    return image, image_cache
 
 
 @spaces.GPU
 def preprocess_sam_image_fn(
-    image: Image.Image,
+    image: Image.Image, req: gr.Request
 ) -> Image.Image:
     if isinstance(image, np.ndarray):
         image = Image.fromarray(image)
 
-    # image.save(f"{TMP_DIR}/{req.session_hash}/raw_image.png")
     sam_image = SAM_PREDICTOR.preprocess_image(image)
+    image_cache = Image.fromarray(sam_image).resize((512, 512))
     SAM_PREDICTOR.predictor.set_image(sam_image)
 
-    return sam_image
+    return sam_image, image_cache
 
 
 def active_btn_by_content(content: gr.Image) -> gr.Button:
@@ -263,6 +250,10 @@ def get_selected_image(
         raise ValueError(f"Invalid choice: {choice}")
 
 
+def get_cached_image(image_path: str) -> Image.Image:
+    return Image.open(image_path).resize((512, 512))
+
+
 @spaces.GPU
 def pack_state(gs: Gaussian, mesh: MeshExtractResult) -> dict:
     return {
@@ -353,6 +344,7 @@ def image_to_3d(
     ss_sampling_steps: int,
     slat_guidance_strength: float,
     slat_sampling_steps: int,
+    raw_image_cache: Image.Image,
     sam_image: Image.Image = None,
     is_sam_image: bool = False,
     req: gr.Request = None,
@@ -361,7 +353,6 @@ def image_to_3d(
         seg_image = filter_image_small_connected_components(sam_image)
         seg_image = Image.fromarray(seg_image, mode="RGBA")
         seg_image = trellis_preprocess(seg_image)
-        # seg_image.save(f"{TMP_DIR}/seg_image_sam.png")
     else:
         seg_image = image
 
@@ -369,6 +360,7 @@ def image_to_3d(
         seg_image = Image.fromarray(seg_image)
 
     seg_image.save(f"{TMP_DIR}/{req.session_hash}/seg_image.png")
+    raw_image_cache.save(f"{TMP_DIR}/{req.session_hash}/raw_image.png")
     PIPELINE.cuda()
     outputs = PIPELINE.run(
         seg_image,
@@ -650,7 +642,7 @@ def text2image_fn(
     if postprocess:
         for idx in range(len(images)):
             image = images[idx]
-            images[idx] = preprocess_image_fn(image)
+            images[idx] = preprocess_image_fn(image, req)
 
     save_paths = []
     for idx, image in enumerate(images):

requirements.txt

@@ -19,7 +19,6 @@ igraph==0.11.8
 pyvista==0.36.1
 openai==1.58.1
 transformers==4.42.4
-# gradio_litmodel3d==0.0.1
 gradio==5.12.0
 sentencepiece==0.2.0
 diffusers==0.31.0