update

app.py

@@ -154,6 +154,13 @@ def vggsfm_predictions_to_glb(predictions):
     rot[:3, :3] = Rotation.from_euler('y', np.deg2rad(180)).as_matrix()
     glbscene.apply_transform(np.linalg.inv(np.linalg.inv(extrinsics_opencv_4x4[0]) @ opengl_mat @ rot))
 
+    # Calculate the bounding box center and apply the translation
+    bounding_box = glbscene.bounds
+    center = (bounding_box[0] + bounding_box[1]) / 2
+    translation = np.eye(4)
+    translation[:3, 3] = -center
+
+    glbscene.apply_transform(translation)
     # glbfile = "glbscene.glb"
     # glbscene.export(file_obj=glbfile)    
     return glbscene
@@ -173,44 +180,100 @@ cake_video = "vggsfm_code/examples/videos/cake_video.mp4"
 cake_images = glob.glob(f'vggsfm_code/examples/cake/images/*')
 british_museum_images = glob.glob(f'vggsfm_code/examples/british_museum/images/*')
 
-if True:
-    demo = gr.Interface(
-        title="🎨 VGGSfM: Visual Geometry Grounded Deep Structure From Motion",
-        fn=vggsfm_demo,
-        inputs=[
-            gr.Video(label="Input video", interactive=True),
-            gr.File(file_count="multiple", label="Input Images", interactive=True),
-            gr.Slider(minimum=1, maximum=10, step=1, value=5, label="Number of query images"),
-            gr.Slider(minimum=512, maximum=4096, step=1, value=1024, label="Number of query points"),
-        ],
-        outputs=[gr.Model3D(label="Reconstruction"), gr.Textbox(label="Log")],
-        examples=[
-            # [apple_video, apple_images, 5, 2048],
-            [cake_video, cake_images, 3, 4096],
-            [british_museum_video, british_museum_images, 2, 4096],],
-        cache_examples=True,
-        # allow_flagging=False,
-        allow_flagging='never',  # Updated from False to 'never'
-        concurrency_limit=1,  # Added concurrency_limit to Interface
-        description = """<div style="text-align: left;"> 
-        <p>Welcome to <a href="https://github.com/facebookresearch/vggsfm" target="_blank">VGGSfM</a> demo! 
-        This space demonstrates 3D reconstruction from input image frames. </p> 
-        <p>To get started quickly, you can click on our examples. If you want to reconstruct your own data, simply: </p> 
-        <ul style="display: inline-block; text-align: left;"> 
-            <li>upload the images (.jpg, .png, etc.), or </li> 
-            <li>upload a video (.mp4, .mov, etc.) </li> 
-        </ul> 
-        <p>If both images and videos are uploaded, the demo will only reconstruct the uploaded images. By default, we extract one image frame per second from the input video. To prevent crashes on the Hugging Face space, we currently limit reconstruction to the first 20 image frames. </p> 
-        <p>For more details, check our <a href="https://github.com/facebookresearch/vggsfm" target="_blank">GitHub Repo</a> ⭐</p> 
-        <p>(Please note that running reconstruction on Hugging Face space is slower than on a local machine.) </p> 
-        </div>""",
+########################################################################################################################
+# if True:
+# demo = gr.Interface(
+#     title="🎨 VGGSfM: Visual Geometry Grounded Deep Structure From Motion",
+#     fn=vggsfm_demo,
+#     inputs=[
+#         gr.Video(label="Input video", interactive=True, scale=1),
+#         gr.File(file_count="multiple", label="Input Images", interactive=True, scale=1),
+#         gr.Slider(minimum=1, maximum=10, step=1, value=5, label="Number of query images", scale=1),
+#         gr.Slider(minimum=512, maximum=4096, step=1, value=1024, label="Number of query points", scale=1),
+#     ],
+#     outputs=[
+#         gr.Model3D(label="Reconstruction", scale=10),
+#         gr.Textbox(label="Log", scale=10)
+#     ],
+#     # outputs=[gr.Model3D(label="Reconstruction", scale=3), gr.Textbox(label="Log", )],
+#     examples=[
+#         # [apple_video, apple_images, 5, 2048],
+#         [cake_video, cake_images, 3, 4096],
+#         [british_museum_video, british_museum_images, 2, 4096],],
+#     cache_examples=False,
+#     # allow_flagging=False,
+#     allow_flagging='never',  # Updated from False to 'never'
+#     concurrency_limit=1,  # Added concurrency_limit to Interface
+#     description = """<div style="text-align: left;"> 
+#     <p>Welcome to <a href="https://github.com/facebookresearch/vggsfm" target="_blank">VGGSfM</a> demo! 
+#     This space demonstrates 3D reconstruction from input image frames. </p> 
+#     <p>To get started quickly, you can click on our examples. If you want to reconstruct your own data, simply: </p> 
+#     <ul style="display: inline-block; text-align: left;"> 
+#         <li>upload the images (.jpg, .png, etc.), or </li> 
+#         <li>upload a video (.mp4, .mov, etc.) </li> 
+#     </ul> 
+#     <p>If both images and videos are uploaded, the demo will only reconstruct the uploaded images. By default, we extract one image frame per second from the input video. To prevent crashes on the Hugging Face space, we currently limit reconstruction to the first 20 image frames. </p> 
+#     <p>For more details, check our <a href="https://github.com/facebookresearch/vggsfm" target="_blank">GitHub Repo</a> ⭐</p> 
+#     <p>(Please note that running reconstruction on Hugging Face space is slower than on a local machine.) </p> 
+#     </div>""",
+# )
+
+with gr.Blocks() as demo:
+    gr.Markdown("# 🎨 VGGSfM: Visual Geometry Grounded Deep Structure From Motion")
+    
+    gr.Markdown("""
+    <div style="text-align: left;"> 
+    <p>Welcome to <a href="https://github.com/facebookresearch/vggsfm" target="_blank">VGGSfM</a> demo! 
+    This space demonstrates 3D reconstruction from input image frames. </p> 
+    <p>To get started quickly, you can click on our examples (page bottom). If you want to reconstruct your own data, simply: </p> 
+    <ul style="display: inline-block; text-align: left;"> 
+        <li>upload the images (.jpg, .png, etc.), or </li> 
+        <li>upload a video (.mp4, .mov, etc.) </li> 
+    </ul> 
+    <p>If both images and videos are uploaded, the demo will only reconstruct the uploaded images. By default, we extract one image frame per second from the input video. To prevent crashes on the Hugging Face space, we currently limit reconstruction to the first 20 image frames. </p> 
+    <p>For more details, check our <a href="https://github.com/facebookresearch/vggsfm" target="_blank">GitHub Repo</a> ⭐</p> 
+    <p>(Please note that running reconstruction on Hugging Face space is slower than on a local machine.) </p> 
+    </div>
+    """)
+
+    with gr.Row():
+        with gr.Column(scale=1):
+            input_video = gr.Video(label="Input video", interactive=True)
+            input_images = gr.File(file_count="multiple", label="Input Images", interactive=True)
+            num_query_images = gr.Slider(minimum=1, maximum=10, step=1, value=5, label="Number of query images")
+            num_query_points = gr.Slider(minimum=512, maximum=4096, step=1, value=1024, label="Number of query points")
+        
+        with gr.Column(scale=3):
+            reconstruction_output = gr.Model3D(label="Reconstruction", height=520)
+            log_output = gr.Textbox(label="Log")
+
+    submit_btn = gr.Button("Submit")
+
+    examples = [
+        [cake_video, cake_images, 3, 4096],
+        [british_museum_video, british_museum_images, 2, 4096],
+    ]
+    
+    gr.Examples(examples=examples, 
+                inputs=[input_video, input_images, num_query_images, num_query_points],
+                outputs=[reconstruction_output, log_output],  # Provide outputs
+                fn=vggsfm_demo,  # Provide the function
+                cache_examples=True
+                )
+
+    submit_btn.click(
+        vggsfm_demo,
+        [input_video, input_images, num_query_images, num_query_points],
+        [reconstruction_output, log_output]
     )
 
     demo.launch(debug=True, share=True)
-else:
-    import glob
-    files = glob.glob(f'vggsfm_code/examples/cake/images/*', recursive=True)
-    vggsfm_demo(files, None, None)
+########################################################################################################################
+
+# else:
+#     import glob
+#     files = glob.glob(f'vggsfm_code/examples/cake/images/*', recursive=True)
+#     vggsfm_demo(files, None, None)
 
     
 # demo.queue(max_size=20, concurrency_count=1).launch(debug=True, share=True)

images_to_videos.py

@@ -2,8 +2,7 @@ import cv2
 import os
 
 # Parameters
-
-name = "cake"
+name = "british_museum"
 folder_path = f'vggsfm_code/examples/{name}/images'  # Update with the path to your images
 video_path = f'{name}_video.mp4'
 fps = 1  # frames per second
@@ -22,7 +21,10 @@ video = cv2.VideoWriter(video_path, fourcc, fps, (width, height))
 
 # Add images to video
 for image in images:
-    video.write(cv2.imread(os.path.join(folder_path, image)))
+    img = cv2.imread(os.path.join(folder_path, image))
+    if img.shape[:2] != (height, width):
+        img = cv2.resize(img, (width, height))  # Resize image to match the first image's size
+    video.write(img)
 
 # Release the video writer
-video.release()
+video.release()