predict script is now working with high dimensional data and writes t…

…iffs in imagej-conform way

n2v/version.py

@@ -1 +1 @@
-__version__ = '0.1.5'
+__version__ = '0.1.6'

scripts/predictN2V.py

@@ -4,10 +4,11 @@
 import sys
 import argparse
 from glob import glob
+import csbdeep.io
 
 parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
 parser.add_argument("--baseDir", help="directory in which all your network will live", default='models')
-parser.add_argument("--name", help="name of your network", default='N2V2D')
+parser.add_argument("--name", help="name of your network", default='N2V')
 parser.add_argument("--dataPath", help="The path to your data")
 parser.add_argument("--fileName", help="name of your data file", default="*.tif")
 parser.add_argument("--output", help="The path to which your data is to be saved", default='.')
@@ -20,6 +21,8 @@
 
 args = parser.parse_args()
 
+print(args.output)
+
 assert (not 'T' in args.dims) or (args.dims[0]=='T')
 
 # We import all our dependencies.
@@ -28,7 +31,6 @@
 import numpy as np
 from matplotlib import pyplot as plt
 from tifffile import imread
-from tifffile import imwrite
 
 
 # A previously trained model is loaded by creating a new N2V-object without providing a 'config'.
@@ -48,31 +50,45 @@
 datagen = N2V_DataGenerator()
 imgs = datagen.load_imgs_from_directory(directory = args.dataPath, dims=args.dims, filter=args.fileName)
 
+
 files = glob(os.path.join(args.dataPath, args.fileName))
 files.sort()
 
 for i, img in enumerate(imgs):
     img_=img
+
+    if 'Z' in args.dims:
+        myDims='TZYXC'
+    else:
+        myDims='TYXC'
+
+    if not 'C' in args.dims :
+        img_=img[...,0]
+        myDims=myDims[:-1]
+
+    myDims_=myDims[1:]
+
+
     if not 'C' in args.dims :
         img_=img[...,0]
 
     # if we have a time dimension we process the images one by one
     if args.dims[0]=='T':
+        outDims=myDims
         pred=img_.copy()
-        myDims=args.dims[1:]	
-
-
         for j in range(img_.shape[0]):
-            pred[j] = model.predict( img_[j], axes=myDims, n_tiles=tiles)
+            print('predicting slice', j, img_[j].shape, myDims_)
+            pred[j] = model.predict( img_[j], axes=myDims_, n_tiles=tiles)
     else:
+        outDims=myDims_
         img_=img_[0,...]
         print("denoising image "+str(i+1) +" of "+str(len(imgs)))
         # Denoise the image.
-        print(args.dims)
-        pred = model.predict( img_, axes=args.dims, n_tiles=tiles)
-
+        pred = model.predict( img_, axes=myDims_, n_tiles=tiles)
+
     print(pred.shape)
     outpath=args.output
     filename=os.path.basename(files[i]).replace('.tif','_N2V.tif')
     outpath=os.path.join(outpath,filename)
-    imwrite(filename,pred.astype(np.float32))
+    print('writing file to ',outpath, outDims, pred.shape)
+    csbdeep.io.save_tiff_imagej_compatible(outpath, pred.astype(np.float32), outDims)

scripts/trainN2V.py

@@ -6,21 +6,21 @@
 
 parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
 parser.add_argument("--baseDir", help="base directory in which your network will live", default='models')
-parser.add_argument("--name", help="name of your network", default='N2V3D')
+parser.add_argument("--name", help="name of your network", default='N2V')
 parser.add_argument("--dataPath", help="The path to your training data")
 parser.add_argument("--fileName", help="name of your training data file", default="*.tif")
-parser.add_argument("--validationFraction", help="Fraction of data you want to use for validation (percent)", default=10.0, type=float)
+parser.add_argument("--validationFraction", help="Fraction of data you want to use for validation (percent)", default=5.0, type=float)
 parser.add_argument("--dims", help="dimensions of your data, can include: X,Y,Z,C (channel), T (time)", default='YX')
 parser.add_argument("--patchSizeXY", help="XY-size of your training patches", default=64, type=int)
 parser.add_argument("--patchSizeZ", help="Z-size of your training patches", default=64, type=int)
 parser.add_argument("--epochs", help="number of training epochs", default=100, type=int)
-parser.add_argument("--stepsPerEpoch", help="number training steps per epoch", default=5, type=int)
+parser.add_argument("--stepsPerEpoch", help="number training steps per epoch", default=400, type=int)
 parser.add_argument("--batchSize", help="size of your training batches", default=64, type=int)
 parser.add_argument("--netDepth", help="depth of your U-Net", default=2, type=int)
 parser.add_argument("--netKernelSize", help="Size of conv. kernels in first layer", default=3, type=int)
 parser.add_argument("--n2vPercPix", help="percentage of pixels to manipulated by N2V", default=1.6, type=float)
 parser.add_argument("--learningRate", help="initial learning rate", default=0.0004, type=float)
-
+parser.add_argument("--unet_n_first", help="number of feature channels in the first u-net layer", default=32, type=int)
 
 if len(sys.argv)==1:
     parser.print_help(sys.stderr)
@@ -83,6 +83,7 @@
                    train_batch_size=args.batchSize, n2v_perc_pix=args.n2vPercPix, n2v_patch_shape=pshape,
                    n2v_manipulator='uniform_withCP', n2v_neighborhood_radius=5, train_learning_rate=args.learningRate,
                    unet_n_depth=args.netDepth,
+                   unet_n_first=args.unet_n_first
                    )
 
 # Let's look at the parameters stored in the config-object.