Implement details of dense depth paper

losses.py

@@ -2,21 +2,19 @@ import tensorflow as tf
 import tensorflow.keras.backend as K
 
 
-def dense_depth_loss_function(y_true, y_pred, theta=0.1, maxDepthVal=1000.0 / 10.0):
-    # Point-wise depth
-    l_depth = K.mean(K.abs(y_pred - y_true), axis=-1)
+def dense_depth_loss_function(y, y_pred):
+    """
+    Implementation of the loss from the dense depth paper https://arxiv.org/pdf/1812.11941.pdf
+    """
+    # Point-wise L1 loss
+    l_depth = K.mean(K.abs(y_pred - y), axis=-1)
 
-    # Edges
-    dy_true, dx_true = tf.image.image_gradients(y_true)
+    # L1 loss over image gradients
+    dy, dx = tf.image.image_gradients(y)
     dy_pred, dx_pred = tf.image.image_gradients(y_pred)
-    l_edges = K.mean(K.abs(dy_pred - dy_true) + K.abs(dx_pred - dx_true), axis=-1)
+    l_grad = K.mean(K.abs(dy_pred - dy) + K.abs(dx_pred - dx), axis=-1)
 
-    # Structural similarity (SSIM) index
-    l_ssim = K.clip((1 - tf.image.ssim(y_true, y_pred, maxDepthVal)) * 0.5, 0, 1)
+    #  Structural Similarity (SSIM)
+    l_ssim = (1 - tf.image.ssim(y, y_pred, 500)) / 2
 
-    # Weights
-    w1 = 1.0
-    w2 = 1.0
-    w3 = theta
-
-    return (w1 * l_ssim) + (w2 * K.mean(l_edges)) + (w3 * K.mean(l_depth))
+    return 0.1 * K.mean(l_depth) + l_grad + l_ssim