Update

README.md

@@ -1,3 +1,13 @@
+<h1>Update</h1>
+
+Pushed a new implementation os starnet in TF2.x. The whole implementation is in one file starnet_v1_TF2.py.
+
+I also created a few Jupyter notebooks for ease of use:
+
+1. starnet_v1_TF2_transform.ipynb - loads and transforms an image.
+2. starnet_v1_TF2.ipynb - more detailed example that loads a model and shows how to train it (really simple as well I think).
+
+
 **StarNet** is a neural network that can remove stars from images in one simple step leaving only background.
 
 More technically it is a convolutional residual net with encoder-decoder architecture and with L1, Adversarial and Perceptual losses.

starnet_v1_TF2.py

@@ -102,38 +102,6 @@ def _ramp(self, x):
         return tf.clip_by_value(x, 0, 1)
 
     def _augmentator(self, o, s):
-        # rotate
-        if np.random.rand() < 0.33:
-            r = np.random.randint(360)
-            pad = int(self.window_size / 2)
-            bc = np.random.rand()
-
-            if np.random.rand() < 0.50:
-                resample = img.BICUBIC
-            elif np.random.rand() < 0.75:
-                resample = img.BILINEAR
-            else:
-                resample = img.NEAREST
-
-            Xtmp = copy.copy(o)
-            Ytmp = copy.copy(s)
-
-            if np.random.rand() < 0.90:
-                Xtmp = np.pad(Xtmp, ((pad, pad), (pad, pad), (0, 0)), mode = 'reflect')
-                Ytmp = np.pad(Ytmp, ((pad, pad), (pad, pad), (0, 0)), mode = 'reflect')
-            else:
-                Xtmp = np.pad(Xtmp, ((pad, pad), (pad, pad), (0, 0)), mode = 'constant', constant_values = bc)
-                Ytmp = np.pad(Ytmp, ((pad, pad), (pad, pad), (0, 0)), mode = 'constant', constant_values = bc)
-
-            Xtmp = img.fromarray(np.uint8(Xtmp * 255))
-            Ytmp = img.fromarray(np.uint8(Ytmp * 255))
-
-            Xtmp = Xtmp.rotate(r, resample = resample)
-            Ytmp = Ytmp.rotate(r, resample = resample)
-
-            o = np.array(Xtmp)[pad:-pad, pad:-pad] / 255
-            s = np.array(Ytmp)[pad:-pad, pad:-pad] / 255
-
         # flip horizontally
         if np.random.rand() < 0.50:
             o = np.flip(o, axis = 1)

wherearemyweights.txt

@@ -1,9 +1,11 @@
-To download pre-trained weights, follow this link (~500Mb)
+To download pre-trained weights, follow the links below:
 
-https://www.dropbox.com/s/atcs42ox4n99w96/starnet_weights.zip?dl=0
+For TF1.x implementation:
 
-And extract all the content into root folder (folder with this file).
+https://www.dropbox.com/s/atcs42ox4n99w96/starnet_weights.zip?dl=0
 
+For TF2.x implementation:
 
+https://www.dropbox.com/s/lcgn5gvnxpo27s5/starnet_weights2.zip?dl=0
 
-Or you might also use lfs (if you know what that means, otherwise ignore this).
+And extract all the content into root folder (folder with this file).