Merge pull request #110 from juglab/tf2

Upgrade to TensorFlow 2
juglab · Jun 24, 2021 · 1c50f5b · 1c50f5b
2 parents ce99376 + ffbd612
commit 1c50f5b
Show file tree

Hide file tree

Showing 23 changed files with 1,081 additions and 656 deletions.
diff --git a/.travis.yml b/.travis.yml
@@ -12,18 +12,10 @@ env:
 
 matrix:
   include:
-    - os: linux
-      dist: bionic
-      python: 3.6
-      env: TENSORFLOW='tensorflow==1.14.0' KERAS='keras==2.2.5'
-    - os: linux
-      dist: bionic
-      python: 3.6
-      env: TENSORFLOW='tensorflow==1.12.0' KERAS='keras==2.2.5'
     - os: linux
       dist: bionic
       python: 3.7
-      env: TENSORFLOW='tensorflow==1.14.0' KERAS='keras==2.2.5'
+      env: TENSORFLOW='tensorflow==2.4.1' KERAS='keras==2.3.1'
 
 install:
   - pip install $TENSORFLOW $KERAS

diff --git a/README.md b/README.md
@@ -13,7 +13,9 @@ Our implementation is based on [CSBDEEP](http://csbdeep.bioimagecomputing.com) (
 
 ## Installation
 This implementation requires [Tensorflow](https://www.tensorflow.org/install/).
-We have tested Noise2Void on LinuxMint 19 and Ubuntu 18.0 using python 3.6 and 3.7 and tensorflow-gpu 1.12.0 and 1.14.0.
+We have tested Noise2Void using Python 3.7 and tensorflow-gpu 2.4.1.
+
+Note: If you want to use TensorFlow 1.15 you have to install N2V v0.2.1. N2V v0.3.0 supports TensorFlow 2 only.
 
 #### If you start from scratch...
 We recommend using [miniconda](https://docs.conda.io/en/latest/miniconda.html).
@@ -22,13 +24,12 @@ If you do not yet have a strong opinion, just use it too!
 After installing Miniconda, the following lines might are likely the easiest way to get Tensorflow and CuDNN installed on your machine (_Note:_ Macs are not supported, and if you sit on a Windows machine all this might also require some modifications.):
 
 ```
-$ conda create -n 'n2v' python=3.6
+$ conda create -n 'n2v' python=3.7
 $ source activate n2v
-$ conda install tensorflow-gpu=1.14 keras=2.2.4
+$ conda install tensorflow-gpu=2.4.1 keras=2.3.1
 $ pip install jupyter
 ```
 
-Note: it is very important that the version of keras be 2.2.4 or 2.2.5, hence the explicit installation above.
 Once this is done (or you had tensorflow et al. installed already), you can install N2V with one of the following two options:
 
 #### Option 1: PIP (current stable release)

diff --git a/examples/2D/denoising2D_BSD68/BSD68_reproducibility.ipynb b/examples/2D/denoising2D_BSD68/BSD68_reproducibility.ipynb
diff --git a/examples/2D/denoising2D_RGB/01_training.ipynb b/examples/2D/denoising2D_RGB/01_training.ipynb
diff --git a/examples/2D/denoising2D_RGB/02_prediction.ipynb b/examples/2D/denoising2D_RGB/02_prediction.ipynb
diff --git a/examples/2D/denoising2D_SEM/01_training.ipynb b/examples/2D/denoising2D_SEM/01_training.ipynb
diff --git a/examples/2D/denoising2D_SEM/02_prediction.ipynb b/examples/2D/denoising2D_SEM/02_prediction.ipynb
diff --git a/examples/2D/structN2V_2D_convallaria/01_training.ipynb b/examples/2D/structN2V_2D_convallaria/01_training.ipynb
diff --git a/examples/2D/structN2V_2D_convallaria/02_prediction.ipynb b/examples/2D/structN2V_2D_convallaria/02_prediction.ipynb
diff --git a/examples/2D/structN2V_2D_synth_mem/train_and_predict.ipynb b/examples/2D/structN2V_2D_synth_mem/train_and_predict.ipynb
diff --git a/examples/3D/01_training.ipynb b/examples/3D/01_training.ipynb
diff --git a/examples/3D/02_prediction.ipynb b/examples/3D/02_prediction.ipynb
diff --git a/n2v/internals/N2V_DataWrapper.py b/n2v/internals/N2V_DataWrapper.py
@@ -1,8 +1,9 @@
-from keras.utils import Sequence
+from csbdeep.internals.train import RollingSequence
+from tensorflow.keras.utils import Sequence
 
 import numpy as np
 
-class N2V_DataWrapper(Sequence):
+class N2V_DataWrapper(RollingSequence):
     """
     The N2V_DataWrapper extracts random sub-patches from the given data and manipulates 'num_pix' pixels in the
     input.
@@ -23,8 +24,9 @@ class N2V_DataWrapper(Sequence):
                         The manipulator used for the pixel replacement.
     """
 
-    def __init__(self, X, Y, batch_size, perc_pix=0.198, shape=(64, 64),
+    def __init__(self, X, Y, batch_size, length, perc_pix=0.198, shape=(64, 64),
                  value_manipulation=None, structN2Vmask=None):
+        super(N2V_DataWrapper, self).__init__(data_size=len(X), batch_size=batch_size, length=length)
         self.X, self.Y = X, Y
         self.batch_size = batch_size
         self.perm = np.random.permutation(len(self.X))
@@ -54,24 +56,23 @@ def __init__(self, X, Y, batch_size, perc_pix=0.198, shape=(64, 64),
         else:
             raise Exception('Dimensionality not supported.')
 
-        self.X_Batches = np.zeros((self.X.shape[0], *self.shape, self.n_chan), dtype=np.float32)
-        self.Y_Batches = np.zeros((self.Y.shape[0], *self.shape, 2*self.n_chan), dtype=np.float32)
-
-    def __len__(self):
-        return int(np.ceil(len(self.X) / float(self.batch_size)))
+        self.X_Batches = np.zeros((self.batch_size, *self.shape, self.n_chan), dtype=np.float32)
+        self.Y_Batches = np.zeros((self.batch_size, *self.shape, 2*self.n_chan), dtype=np.float32)
 
     def on_epoch_end(self):
         self.perm = np.random.permutation(len(self.X))
-        self.X_Batches *= 0
-        self.Y_Batches *= 0
+
 
     def __getitem__(self, i):
-        idx = slice(i * self.batch_size, (i + 1) * self.batch_size)
-        idx = self.perm[idx]
+        idx = self.batch(i)
+        # idx = slice(i * self.batch_size, (i + 1) * self.batch_size)
+        # idx = self.perm[idx]
+        self.X_Batches *= 0
+        self.Y_Batches *= 0
         self.patch_sampler(self.X, self.X_Batches, indices=idx, range=self.range, shape=self.shape)
 
         for c in range(self.n_chan):
-            for j in idx:
+            for j in range(self.batch_size):
                 coords = self.get_stratified_coords(self.rand_float, box_size=self.box_size,
                                                     shape=self.shape)
 
@@ -87,7 +88,7 @@ def __getitem__(self, i):
                 if self.structN2Vmask is not None:
                     self.apply_structN2Vmask(self.X_Batches[j, ..., c], coords, self.dims, self.structN2Vmask)
 
-        return self.X_Batches[idx], self.Y_Batches[idx]
+        return self.X_Batches, self.Y_Batches
 
     def apply_structN2Vmask(self, patch, coords, dims, mask):
         """
@@ -112,18 +113,18 @@ def apply_structN2Vmask(self, patch, coords, dims, mask):
     # return x_val_structN2V, indexing_structN2V
     @staticmethod
     def __subpatch_sampling2D__(X, X_Batches, indices, range, shape):
-        for j in indices:
+        for i, j in enumerate(indices):
             y_start = np.random.randint(0, range[0] + 1)
             x_start = np.random.randint(0, range[1] + 1)
-            X_Batches[j] = np.copy(X[j, y_start:y_start + shape[0], x_start:x_start + shape[1]])
+            X_Batches[i] = np.copy(X[j, y_start:y_start + shape[0], x_start:x_start + shape[1]])
 
     @staticmethod
     def __subpatch_sampling3D__(X, X_Batches, indices, range, shape):
-        for j in indices:
+        for i, j in enumerate(indices):
             z_start = np.random.randint(0, range[0] + 1)
             y_start = np.random.randint(0, range[1] + 1)
             x_start = np.random.randint(0, range[2] + 1)
-            X_Batches[j] = np.copy(X[j, z_start:z_start + shape[0], y_start:y_start + shape[1], x_start:x_start + shape[2]])
+            X_Batches[i] = np.copy(X[j, z_start:z_start + shape[0], y_start:y_start + shape[1], x_start:x_start + shape[2]])
 
     @staticmethod
     def __get_stratified_coords2D__(coord_gen, box_size, shape):

diff --git a/n2v/internals/n2v_losses.py b/n2v/internals/n2v_losses.py
@@ -1,4 +1,4 @@
-import keras.backend as K
+import tensorflow.keras.backend as K
 
 import tensorflow as tf
 

diff --git a/n2v/models/n2v_config.py b/n2v/models/n2v_config.py
@@ -1,14 +1,12 @@
 import argparse
 
-import keras.backend as K
+import tensorflow.keras.backend as K
 
-from csbdeep.models import BaseConfig
 from csbdeep.utils import _raise, axes_check_and_normalize, axes_dict, backend_channels_last
 
 from six import string_types
 
 import numpy as np
-from logging.config import BaseConfigurator
 
 # This class is a adapted version of csbdeep.models.config.py.
 class N2VConfig(argparse.Namespace):