1

app.py

@@ -116,6 +116,7 @@ def generate_monocular_depth_maps(img_list, depth_prior_name):
           depth = pipe(image)["predicted_depth"].numpy()
           depth = cv2.resize(depth[0], image.size, interpolation=cv2.INTER_LANCZOS4)
           focallength_px = 200
+          print(depth.max(),depth.min())
           depth_list.append(depth)
           focallength_px_list.append(focallength_px)
           #np.savez_compressed(path_depthanything, depth=depth)  
@@ -138,6 +139,7 @@ def local_get_reconstructed_scene(filelist, min_conf_thr, as_pointcloud, mask_sk
     model = AsymmetricCroCo3DStereo.from_pretrained(weights_path).to(device)  
     output = inference(pairs, model, device, batch_size=batch_size, verbose=not silent)
     mode = GlobalAlignerMode.PointCloudOptimizer  
+    print(output)
     scene = global_aligner(output, device=device, mode=mode, verbose=not silent, shared_focal = True, temporal_smoothing_weight=0.01, translation_weight=1.0,
                                flow_loss_weight=0.01, flow_loss_start_epoch=0.1, flow_loss_thre=25, use_self_mask=True,
                                num_total_iter=300, empty_cache= len(filelist) > 72)
@@ -192,13 +194,6 @@ with gradio.Blocks(css=css, title=title, delete_cache=(gradio_delete_cache, grad
                      [os.path.join(HERE_PATH, 'example/bear/00000.jpg'),
                       os.path.join(HERE_PATH, 'example/bear/00001.jpg'),
                       os.path.join(HERE_PATH, 'example/bear/00002.jpg'),
-                      os.path.join(HERE_PATH, 'example/bear/00003.jpg'),
-                      os.path.join(HERE_PATH, 'example/bear/00004.jpg'),
-                      os.path.join(HERE_PATH, 'example/bear/00005.jpg'),
-                      os.path.join(HERE_PATH, 'example/bear/00006.jpg'),
-                      os.path.join(HERE_PATH, 'example/bear/00007.jpg'),
-                      os.path.join(HERE_PATH, 'example/bear/00008.jpg'),
-                      os.path.join(HERE_PATH, 'example/bear/00009.jpg'),
                       ]
                 ],
                 [

third_party/RAFT/core/extractor.py

@@ -312,7 +312,7 @@ class ResNetFPN(nn.Module):
                     nn.init.constant_(m.weight, 1)
                 if m.bias is not None:
                     nn.init.constant_(m.bias, 0)
-
+        #print('****',args.pretrain, self.init_weight)
         if self.init_weight:
             from torchvision.models import resnet18, ResNet18_Weights, resnet34, ResNet34_Weights
             if args.pretrain == 'resnet18':