Update app.py

app.py

@@ -53,8 +53,6 @@ set_seed(666)
 
 DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
-# DEVICE = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
-# DEVICE = 'cuda' if torch.cuda.is_available() else 'cpu'
 default_steps = 20
 model = PixelPerfectDepth(sampling_steps=default_steps)
 ckpt_path = hf_hub_download(
@@ -82,19 +80,13 @@ def main(share=True):
 
     @(spaces.GPU if HUGGINFACE_SPACES_INSTALLED else (lambda x: x))
     def predict_depth(image, denoise_steps):
-        # DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-        # global model
-        # model = model.to(DEVICE)
         depth, resize_image = model.infer_image(image, sampling_steps=denoise_steps)
         return depth, resize_image
 
     @(spaces.GPU if HUGGINFACE_SPACES_INSTALLED else (lambda x: x))
     def predict_moge_depth(image):
-        # DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
         image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
         image = torch.tensor(image / 255, dtype=torch.float32, device=DEVICE).permute(2, 0, 1)
-        # global moge_model
-        # moge_model = moge_model.to(DEVICE)
         metric_depth, mask, intrinsics = moge_model.infer(image)
         metric_depth[~mask] = metric_depth[mask].max()
         return metric_depth, mask, intrinsics
@@ -102,9 +94,7 @@ def main(share=True):
     def on_submit(image, denoise_steps, apply_filter, request: gr.Request = None):
 
         H, W = image.shape[:2]
-
         ppd_depth, resize_image = predict_depth(image[:, :, ::-1], denoise_steps)
-
         resize_H, resize_W = resize_image.shape[:2]
 
         # moge provide metric depth and intrinsics
@@ -132,12 +122,15 @@ def main(share=True):
         ply_path = os.path.join(output_path, 'pointcloud.ply')
 
         # save pcd to temporary .ply
+        pcd.points = o3d.utility.Vector3dVector(
+            np.asarray(pcd.points) * np.array([1, -1, -1], dtype=np.float32)
+        )
         o3d.io.write_point_cloud(ply_path, pcd)
-        
         vertices = np.asarray(pcd.points)
         vertex_colors = (np.asarray(pcd.colors) * 255).astype(np.uint8)
-        vertices[:, 2] = -vertices[:, 2]
-        vertices[:, 1] = -vertices[:, 1]
+        # vertices = vertices * np.array([1, -1, -1], dtype=np.float32)
+        # vertices[:, 2] = -vertices[:, 2]
+        # vertices[:, 1] = -vertices[:, 1]
         mesh = trimesh.PointCloud(vertices=vertices, colors=vertex_colors)
         glb_path = os.path.join(output_path, 'pointcloud.glb')
         mesh.export(glb_path)
@@ -172,7 +165,7 @@ def main(share=True):
     with gr.Blocks(theme=gr.themes.Soft()) as demo:
         gr.Markdown(title)
         gr.Markdown(description)
-        gr.Markdown("### Depth Prediction demo")
+        gr.Markdown("### Point Cloud & Depth Prediction demo")
 
         with gr.Row():
             # Left: input image + settings