init project

app.py

@@ -39,8 +39,6 @@ import torchvision.transforms as tvf
 
 
 silent = False
-pe3r = Models('cuda' if torch.cuda.is_available() else 'cpu')
-
 
 def _convert_scene_output_to_glb(outdir, imgs, pts3d, mask, focals, cams2world, cam_size=0.05,
                                  cam_color=None, as_pointcloud=False,
@@ -244,6 +242,7 @@ def slerp_multiple(vectors, t_values):
     return interpolated_vector
 
 @torch.no_grad
+@spaces.GPU(duration=180)
 def get_mask_from_img_sam1(mobilesamv2, yolov8, sam1_image, yolov8_image, original_size, input_size, transform):
     device = 'cuda' if torch.cuda.is_available() else 'cpu'
 
@@ -298,9 +297,24 @@ def get_mask_from_img_sam1(mobilesamv2, yolov8, sam1_image, yolov8_image, origin
 
     return ret_mask
 
-@torch.no_grad
-def get_cog_feats(images):
+@spaces.GPU(duration=180)
+def get_reconstructed_scene(outdir, filelist, schedule, niter, min_conf_thr,
+                            as_pointcloud, mask_sky, clean_depth, transparent_cams, cam_size,
+                            scenegraph_type, winsize, refid):
+    """
+    from a list of images, run dust3r inference, global aligner.
+    then run get_3D_model_from_scene
+    """
+    if len(filelist) < 2:
+        raise gradio.Error("Please input at least 2 images.")
+    
     device = 'cuda' if torch.cuda.is_available() else 'cpu'
+
+    pe3r = Models(device)
+
+    images = Images(filelist=filelist, device=device)
+    
+    # try:
     cog_seg_maps = []
     rev_cog_seg_maps = []
     inference_state = pe3r.sam2.init_state(images=images.sam2_images, video_height=images.sam2_video_size[0], video_width=images.sam2_video_size[1])
@@ -433,25 +447,8 @@ def get_cog_feats(images):
             multi_view_clip_feats[i] = torch.zeros((1024))
     multi_view_clip_feats[mask_num] = torch.zeros((1024))
 
-    return cog_seg_maps, rev_cog_seg_maps, multi_view_clip_feats
+    cog_feats = multi_view_clip_feats
 
-@spaces.GPU(duration=180)
-def get_reconstructed_scene(outdir, filelist, schedule, niter, min_conf_thr,
-                            as_pointcloud, mask_sky, clean_depth, transparent_cams, cam_size,
-                            scenegraph_type, winsize, refid):
-    """
-    from a list of images, run dust3r inference, global aligner.
-    then run get_3D_model_from_scene
-    """
-    if len(filelist) < 2:
-        raise gradio.Error("Please input at least 2 images.")
-    
-    device = 'cuda' if torch.cuda.is_available() else 'cpu'
-
-    images = Images(filelist=filelist, device=device)
-    
-    # try:
-    cog_seg_maps, rev_cog_seg_maps, cog_feats = get_cog_feats(images)
     imgs = load_images(images, rev_cog_seg_maps, size=512, verbose=not silent)
     # except Exception as e:
         # rev_cog_seg_maps = []