init

app.py

@@ -136,7 +136,7 @@ def camera_understanding(image_src, question, seed, progress=gr.Progress(track_t
             suffix = f"_{k}"
         out_path = os.path.join(save_dir, f"camera_map_vis{suffix}.png")
         plt.tight_layout()
-        fig.savefig(out_path, dpi=200, bbox_inches='tight', pad_inches=0)
+        fig.savefig(out_path, dpi=100, bbox_inches='tight', pad_inches=0)
         plt.close(fig)
         saved_paths.append(out_path)
 
@@ -146,6 +146,9 @@ def camera_understanding(image_src, question, seed, progress=gr.Progress(track_t
     w, h = img_up.size
     left = max(0, w - h)
     img_up = img_up.crop((left, 0, w, h))
+
+    w, h = img_lat.size
+    left = max(0, w - h)
     img_lat = img_lat.crop((left, 0, w, h))
 
     return text, [img_up], [img_lat]
@@ -213,9 +216,9 @@ with gr.Blocks(css=css) as demo:
     with gr.Tab("Camera-controllable Image Generation"):
         gr.Markdown(value="## Camera-controllable Image Generation")
 
-        prompt_input = gr.Textbox(label="Prompt.")
+        prompt_input = gr.Textbox(label="Scene Prompt")
 
-        with gr.Accordion("Camera Parameters", open=True):
+        with gr.Accordion("Camera Parameters (in radius)", open=True):
             with gr.Row():
                 roll = gr.Slider(minimum=-0.7854, maximum=0.7854, value=0.1000, step=0.1000, label="roll value")
                 pitch = gr.Slider(minimum=-0.7854, maximum=0.7854, value=-0.1000, step=0.1000, label="pitch value")