Add pixel loss functions, move warp

unsupervised/train.py

@@ -4,7 +4,6 @@ Trainer to learn depth information on unlabeled data (raw images/videos)
 Allows pluggable depth networks for differing performance (including fast-depth)
 """
 
-import tensorflow as tf
 import tensorflow.keras as keras
 
 
@@ -17,23 +16,5 @@ class SFMLearner(keras.Model):
         pass
 
 
-def projective_inverse_warp(depth, pose, t_img, s_imgs, intrinsics):
-    '''
-    SFM Learner inverse warp step
-    ps ~ K.T(t->s).Dt(pt).K^-1.pt
-
-    projected source pixel 
-    '''
-    pass
-
-
-def bilinear_sample(projected_coords, s_img):
-    '''
-    Sample the 4 closest pixels in the source image via the projected coordinates
-    to get the source image warped to the target image
-    '''
-    pass
-
-
 def make_sfm_learner_pose_net(input_shape=(224, 224, 3)):
     pass

unsupervised/warp.py

@@ -0,0 +1,15 @@
+def projective_inverse_warp(target_img, source_img, depth, pose, intrinsics):
+    """
+    Calculate the reprojected image from the source to the target, based on the given depth, pose and intrinsics
+
+    SFM Learner inverse warp step
+        ps ~ K.T(t->s).Dt(pt).K^-1.pt
+
+    :param target_img: Tensor (batch, height, width, 3)
+    :param source_img: Tensor, same shape as target_img
+    :param depth: Tensor, (batch, height, width, 1)
+    :param pose: (batch, 3, 3)
+    :param intrinsics: (batch, 3, 3)
+    :return: The source image reprojected to the target
+    """
+    pass