Let's see.

app.py

@@ -1,16 +1,54 @@
 from depth import MidasDepth
 import gradio as gr
 import numpy as np
-import cv2
+import tempfile
 
 
 depth_estimator = MidasDepth()
 
 
 def get_depth(rgb):
+    rgb = rgb.convert("RGB")
     depth = depth_estimator.get_depth(rgb)
 
-    return rgb, (depth.clip(0, 64) * 1024).astype("uint16")
+    h, w, _ = rgb.shape
+    grid = np.mgrid[0:h, 0:w].transpose(1, 2, 0
+                                        ).reshape(-1, 2)[..., ::-1]
+    flat_grid = grid[:, 1] * w + grid[:, 0]
+
+    positions = np.concatenate(((grid - np.array([[w, h]])
+                                 / 2) / w * 2,
+                                depth.flatten()[flat_grid][..., np.newaxis]),
+                               axis=-1)
+    positions[:, :-1] *= positions[:, -1:]
+    positions[:, :2] *= -1
+
+    pick_edges = depth < 0
+    y, x = (t.flatten() for t in np.mgrid[0:h, 0:w])
+    faces = np.concatenate((
+        np.stack((y * w + x,
+                  (y - 1) * w + x,
+                  y * w + (x - 1)), axis=-1)
+        [(~pick_edges.flatten()) * (x > 0) * (y > 0)],
+        np.stack((y * w + x,
+                  (y + 1) * w + x,
+                  y * w + (x + 1)), axis=-1)
+        [(~pick_edges.flatten()) * (x < w - 1) * (y < im.shape[0] - 1)]
+    ))
+
+    tf = tempfile.NamedTemporaryFile(suffix=".obj").name
+    save_obj(positions, rgb.reshape(-1, 3), faces, tf)
+
+    return rgb, (depth.clip(0, 64) * 1024).astype("uint16"), tf
+
+
+def save_obj(positions, rgb, faces, filename):
+    with open(filename, "w") as f:
+        for position, color in zip(positions, rgb):
+            f.write(
+                f"v {' '.join(map(str, position))} {' '.join(map(str, color))}")
+        for face in faces:
+            f.write(f"f {' '.join(map(str, face))}")
 
 
 gr.Interface(fn=get_depth, inputs=[
@@ -18,5 +56,6 @@ gr.Interface(fn=get_depth, inputs=[
 ], outputs=[
     gr.components.Image(type="pil", label="image"),
     gr.components.Image(type="numpy", label="depth"),
+    gr.components.Model3D()
 
 ]).launch(share=True)

requirements.txt

@@ -1,3 +1,2 @@
 torch
-opencv-python
 timm