result visualization and Readme update

README.md

@@ -10,31 +10,31 @@ There are a few dependencies required to run the code.  They are listed below:
 
 ### System Environments:
 
-**Python 3.5+**
+`Python >= 3.5`
 
-**CUDA** **10.1**
+`CUDA 10.1`
 
-**CuDnn**
+`CuDnn`
 
 ### Pip dependencies:
 
-**Pytorch 1.5.0**
+`Pytorch 1.5.0`
 
-**Torchvision 0.6.0**
+`Torchvision 0.6.0`
 
-**Kornia**
+`Kornia`
 
-**Graphviz**
+`Graphviz`
 
-**Opencv**
+`Opencv`
 
-**Scipy**
+`Scipy`
 
-**Matplotlib**
+`Matplotlib`
 
-**Pandas**
+`Pandas`
 
-**TensorboardX**
+`TensorboardX`
 
 
 
@@ -132,4 +132,24 @@ Similarly, there are several options that you can choose when running validation
 
 
 
-## **Finally, HAVE FUN with the CODE!!!!**
+### Detecting and Infering:
+
+This repository also provided a single pair detecting script so that you can see the result of  DPCN directly.  A few demos are given in the "./demo" directory including images pairs from both simulation dataset and Aero-Ground Dataset. You could customize the script `detect.py` to test the chosen images pair with relative pretrained model given, and run this code below:
+
+`python detect.py`
+
+The results should be something like this:
+
+
+
+![](./images_for_readme/Result1.png)
+
+![Result2](./images_for_readme/Result2.png)
+
+
+
+
+
+
+
+## HAVE FUN with the CODE!!!!

detect.py

@@ -61,9 +61,9 @@ def detect_model(template_path, source_path, model_template, model_source, model
 
 
 
-checkpoint_path = "./checkpoints/laser_sat_qsdjt_9epoch.pt"
-template_path = "./demo/temp_2.jpg"
-source_path = "./demo/src_2.jpg"
+checkpoint_path = "./checkpoints/checkpoint_simulation_hetero.pt"
+template_path = "./demo/temp_1.png"
+source_path = "./demo/src_1.png"
 
 load_pretrained =True
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

utils/utils.py

@@ -202,7 +202,7 @@ def align_image(template, source):
     source = source.cpu().detach().numpy()
     template = template.squeeze(0)  # remove the fake batch dimension
     source = source.squeeze(0)  # remove the fake batch dimension
-    dst = cv2.addWeighted(template, 0.4, source, 0.6, 0)
+    dst = cv2.addWeighted(template, 1, source, 0.6, 0)
     # plt.imshow(dst, cmap="gray")
     # plt.show()  # pause a bit so that plots are updated
     return dst