docs, fix resize

examples/GAN/Image2Image.py

@@ -14,17 +14,21 @@ from tensorpack import *
 from tensorpack.utils.viz import *
 from tensorpack.tfutils.summary import add_moving_summary, summary_moving_average
 import tensorpack.tfutils.symbolic_functions as symbf
-from GAN import GANTrainer, RandomZData, build_GAN_losses
+from GAN import GANTrainer, build_GAN_losses
 
 """
 To train:
     ./Image2Image.py --data /path/to/datadir --mode {AtoB,BtoA}
-    # datadir should contain images of shpae 2s x s, formed by A and B
-    # you can download some data from the original pix2pix repo: https://github.com/phillipi/pix2pix#datasets
+    # datadir should contain jpg images of shpae 2s x s, formed by A and B
+    # you can download some data from the original authors: https://people.eecs.berkeley.edu/~tinghuiz/projects/pix2pix/datasets/
     # training visualization will appear be in tensorboard
 
+Speed:
+    On GTX1080 with BATCH=1, the speed is about 9.3it/s (the original torch version is 9.5it/s)
+
 To visualize on test set:
-    ./Image2Image.py --sample --data /path/to/test/datadir --mode {AtoB,BtoA} --load pretrained.model
+    ./Image2Image.py --sample --data /path/to/test/datadir --mode {AtoB,BtoA} --load model
+
 """
 
 SHAPE = 256
@@ -41,7 +45,7 @@ class Model(ModelDesc):
 
     def generator(self, imgs):
         # imgs: input: 256x256xch
-        # U-Net structure, slightly different from the original on the location of relu/lrelu
+        # U-Net structure, it's slightly different from the original on the location of relu/lrelu
         with argscope(BatchNorm, use_local_stat=True), \
                 argscope(Dropout, is_training=True):
             # always use local stat for BN, and apply dropout even in testing
@@ -118,7 +122,7 @@ class Model(ModelDesc):
             fake_output = tf.image.grayscale_to_rgb(fake_output)
         viz = (tf.concat(2, [input, output, fake_output]) + 1.0) * 128.0
         viz = tf.cast(tf.clip_by_value(viz, 0, 255), tf.uint8, name='viz')
-        tf.summary.image('gen', viz, max_outputs=max(30, BATCH))
+        tf.summary.image('input,output,fake', viz, max_outputs=max(30, BATCH))
 
         all_vars = tf.trainable_variables()
         self.g_vars = [v for v in all_vars if v.name.startswith('gen/')]
@@ -134,9 +138,9 @@ def split_input(img):
     if args.mode == 'BtoA':
         input, output = output, input
     if IN_CH == 1:
-        input = cv2.cvtColor(input, cv2.COLOR_RGB2GRAY)
+        input = cv2.cvtColor(input, cv2.COLOR_RGB2GRAY)[:,:,np.newaxis]
     if OUT_CH == 1:
-        output = cv2.cvtColor(output, cv2.COLOR_RGB2GRAY)
+        output = cv2.cvtColor(output, cv2.COLOR_RGB2GRAY)[:,:,np.newaxis]
     return [input, output]
 
 def get_data():

examples/GAN/README.md

@@ -8,7 +8,7 @@ Reproduce the following GAN-related papers:
 
 + InfoGAN: Interpretable Representation Learning by Information Maximizing GAN. [paper](https://arxiv.org/abs/1606.03657)
 
-Detailed usage is in the docstring of each script.
+Please see the __docstring__ in each script for detailed usage.
 
 ## DCGAN-CelebA.py
 
@@ -27,7 +27,6 @@ Play with the [pretrained model](https://drive.google.com/drive/folders/0B9IPQTv
 ## Image2Image.py
 
 Image-to-Image following the setup in [pix2pix](https://github.com/phillipi/pix2pix).
-It requires the datasets released by the original authors.
 
 For example, with the cityscapes dataset, it learns to generate semantic segmentation map of urban scene:
 

tensorpack/dataflow/image.py

@@ -19,7 +19,7 @@ class ImageFromFile(RNGDataFlow):
         :param channel: 1 or 3 channel
         :param resize: a (h, w) tuple. If given, will force a resize
         """
-        assert len(files), "No Image Files!"
+        assert len(files), "No image files given to ImageFromFile!"
         self.files = files
         self.channel = int(channel)
         self.imread_mode = cv2.IMREAD_GRAYSCALE if self.channel == 1 else cv2.IMREAD_COLOR

tensorpack/dataflow/imgaug/noname.py

@@ -55,9 +55,12 @@ class Resize(ImageAugmentor):
         self._init(locals())
 
     def _augment(self, img, _):
-        return cv2.resize(
+        ret = cv2.resize(
             img, self.shape[::-1],
             interpolation=self.interp)
+        if img.ndim == 3 and ret.ndim == 2:
+            ret = ret[:,:,np.newaxis]
+        return ret
 
 class ResizeShortestEdge(ImageAugmentor):
     """ Resize the shortest edge to a certain number while
@@ -71,8 +74,10 @@ class ResizeShortestEdge(ImageAugmentor):
         h, w = img.shape[:2]
         scale = self.size / min(h, w)
         desSize = map(int, [scale * w, scale * h])
-        img = cv2.resize(img, tuple(desSize), interpolation=cv2.INTER_CUBIC)
-        return img
+        ret = cv2.resize(img, tuple(desSize), interpolation=cv2.INTER_CUBIC)
+        if img.ndim == 3 and ret.ndim == 2:
+            ret = ret[:,:,np.newaxis]
+        return ret
 
 class RandomResize(ImageAugmentor):
     """ randomly rescale w and h of the image"""
@@ -105,5 +110,8 @@ class RandomResize(ImageAugmentor):
                 return img.shape[1], img.shape[0]
 
     def _augment(self, img, dsize):
-        return cv2.resize(img, dsize, interpolation=self.interp)
+        ret = cv2.resize(img, dsize, interpolation=self.interp)
+        if img.ndim == 3 and ret.ndim == 2:
+            ret = ret[:,:,np.newaxis]
+        return ret