update cam convention

ashawkey · Dec 1, 2023 · 7b1f675 · 7b1f675
1 parent b895667
commit 7b1f675
Show file tree

Hide file tree

Showing 2 changed files with 30 additions and 5 deletions.
diff --git a/kiui/cam.py b/kiui/cam.py
@@ -8,6 +8,31 @@
 # elevation in (-90, 90), from +y (-90) --> -y (+90)
 # azimuth in (-180, 180), from +z (0/-360) --> +x (90/-270) --> -z (180/-180) --> -x (270/-90) --> +z (360/0)
 
+''' common world coordinate system conventions
+
+   OpenGL          OpenCV           Blender        Unity             
+Right-handed       Colmap                        Left-handed  
+
+     +y                +z           +z  +y         +y  +z                                               
+     |                /             |  /           |  /                                               
+     |               /              | /            | /                                                   
+     |______+x      /______+x       |/_____+x      |/_____+x                                          
+    /               |                                                                                        
+   /                |                                                                                                  
+  /                 |                                                                                         
+ +z                 +y                                                                                           
+
+'''
+
+''' camera pose matrix
+[[Forward_x, Up_x, Right_x, Position_x],
+ [Forward_y, Up_y, Right_y, Position_y],
+ [Forward_z, Up_z, Right_z, Position_z],
+ [0,         0,    0,       1         ]]
+The xyz follows corresponding world coordinate system.
+'''
+
+
 # look at rotation matrix
 def look_at(campos, target, opengl=True):
     # campos: [N, 3], camera/eye position

diff --git a/kiui/cli/aes.py b/kiui/cli/aes.py
@@ -19,7 +19,7 @@ def __init__(self, device, model_name='openai/clip-vit-large-patch14'):
         self.processor = CLIPProcessor.from_pretrained(model_name)
 
     def encode_image(self, image):
-        # image: PIL, np.ndarray uint8 [H, W, 3]
+        # image: PIL, np.ndarray uint8 [H, W, 3] or [B, H, W, 3]
 
         pixel_values = self.processor(images=image, return_tensors="pt").pixel_values.to(self.device)
         image_features = self.clip_model.get_image_features(pixel_values=pixel_values)
@@ -66,13 +66,13 @@ def __init__(self, device):
 
 
     def __call__(self, x):
-        # x: np.ndarray, (h, w, 3), uint8, RGB
+        # x: np.ndarray, (h, w, 3) / (b, h, w, 3), uint8, RGB
         # return: y: aesthetic score
 
         features = self.clip.encode_image(x)
         y = self.mlp(features)
 
-        return y.item()
+        return y
 
 
 if __name__ == '__main__':
@@ -98,5 +98,5 @@ def __call__(self, x):
     elif image.shape[-1] == 1:
         image = np.concatenate([image] * 3, axis=-1)
 
-    aes_score = aes(image)
-    print(f'Aesthetic score: {aes_score:.3f} <-- {opt.image}')
+    aes_score = aes(image[None, :])
+    print(f'Aesthetic score: {aes_score} <-- {opt.image}')