[pls merge] openpose example (#787)

* tl upadte of Openpose

* code style improvement

* test

* yapf

* comments

* openpose example

* openpose data augmentation API

* train function update

* update non mask case

* add args.parser

* add args.parser for log

* add inference

* paf process added

* paf install instrcution

* add inference

* update md

* hao modified

* remove useless files

* hao arranged files

* trainable

* del

* release prepro for keypoints / vgg19 break

* writing inferencing

* coco 2017

* TODO distrubuted training

* distributed code in same file

* add int wout hout

* agg

* train mode

* add cpm between vgg and stage1

* used own defined 10 layers of VGG19

* remove generated files (#801)

* add mode;

* remove openpose from current repo

* Update CHANGELOG.md

* add keypoint docs;

* yapf - prepro

* fix yapf

* yapf

Author: zsdonghao

CHANGELOG.md

@@ -81,6 +81,7 @@ To release a new version, please update the changelog as followed:
   - Add `tl.files.load_graph_` (PR ＃751)
   - Add `tl.files.save_graph_and_params` (PR ＃751)
   - Add `tl.files.load_graph_and_params` (PR ＃751)
+  - Add `tl.prepro.keypoint_random_xxx` (PR #787)
 - Documentation:
   - Add binary, ternary and dorefa links (PR #711)
   - Update input scale of VGG16 and VGG19 to 0~1 (PR #736)
@@ -152,6 +153,7 @@ To release a new version, please update the changelog as followed:
 - @mutewall: #735
 - @thangvubk: #759
 - @JunbinWang: #796
+- @boldjoel: #787
 
 ## [1.9.1] - 2018-07-30
 

docs/modules/prepro.rst

@@ -87,6 +87,13 @@ Some of the code in this package are borrowed from Keras.
    obj_box_shift
    obj_box_zoom
 
+   keypoint_random_crop
+   keypoint_random_crop2
+   keypoint_random_rotate
+   keypoint_random_flip
+   keypoint_random_resize
+   keypoint_random_resize_shortestedge
+
    pad_sequences
    remove_pad_sequences
    process_sequences
@@ -416,8 +423,30 @@ Image Aug - Zoom
 ^^^^^^^^^^^^^^^^^^^^^^^^^
 .. autofunction:: obj_box_zoom
 
+Keypoints
+------------
 
+Image Aug - Crop
+^^^^^^^^^^^^^^^^^^^^
+.. autofunction:: keypoint_random_crop
+
+.. autofunction:: keypoint_random_crop2
+
+Image Aug - Rotate
+^^^^^^^^^^^^^^^^^^^^
+.. autofunction:: keypoint_random_rotate
+
+Image Aug - Flip
+^^^^^^^^^^^^^^^^^^^^
+.. autofunction:: keypoint_random_flip
+
+Image Aug - Resize
+^^^^^^^^^^^^^^^^^^^^
+.. autofunction:: keypoint_random_resize
 
+Image Aug - Resize Shortest Edge
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+.. autofunction:: keypoint_random_resize_shortestedge
 
 
 Sequence

example/tutorial_models_vgg19.py

@@ -3,6 +3,9 @@
 """VGG-19 for ImageNet using TL models."""
 
 import time
+import skimage
+import skimage.io
+import skimage.transform
 import numpy as np
 import tensorflow as tf
 import tensorlayer as tl
@@ -24,17 +27,37 @@
 
 vgg.print_layers()
 
-img1 = tl.vis.read_image('data/tiger.jpeg')
-img1 = tl.prepro.imresize(img1, (224, 224))
+
+# img1 = tl.vis.read_image('data/tiger.jpeg')
+# img1 = tl.prepro.imresize(img1, (224, 224))
+def load_image(path):
+    # load image
+    img = skimage.io.imread(path)
+    img = img / 255.0
+    if ((0 <= img).all() and (img <= 1.0).all()) is False:
+        raise Exception("image value should be [0, 1]")
+    # print "Original Image Shape: ", img.shape
+    # we crop image from center
+    short_edge = min(img.shape[:2])
+    yy = int((img.shape[0] - short_edge) / 2)
+    xx = int((img.shape[1] - short_edge) / 2)
+    crop_img = img[yy:yy + short_edge, xx:xx + short_edge]
+    # resize to 224, 224
+    resized_img = skimage.transform.resize(crop_img, (224, 224), anti_aliasing=False)
+    return resized_img
+
+
+img1 = load_image("data/tiger.jpeg")  # test data in github
+img1 = img1.reshape((1, 224, 224, 3))
 
 # rescale pixels values in the range of 0-1
-img1 = img1 / 255.0
+# img1 = img1 / 255.0
 if ((0 <= img1).all() and (img1 <= 1.0).all()) is False:
     raise Exception("image value should be [0, 1]")
 
-_ = sess.run(probs, feed_dict={x: [img1]})[0]  # 1st time takes time to compile
+_ = sess.run(probs, feed_dict={x: img1})[0]  # 1st time takes time to compile
 start_time = time.time()
-prob = sess.run(probs, feed_dict={x: [img1]})[0]
+prob = sess.run(probs, feed_dict={x: img1})[0]
 print("  End time : %.5ss" % (time.time() - start_time))
 preds = (np.argsort(prob)[::-1])[0:5]
 for p in preds:

example/tutorial_vgg19.py

@@ -233,6 +233,7 @@ def Vgg19_simple_api(rgb):
 
 img1 = load_image("data/tiger.jpeg")  # test data in github
 img1 = img1.reshape((1, 224, 224, 3))
+
 start_time = time.time()
 prob = sess.run(probs, feed_dict={x: img1})
 print("End time : %.5ss" % (time.time() - start_time))

tensorlayer/files/utils.py

@@ -2277,6 +2277,7 @@ def _dlProgress(count, blockSize, totalSize, pbar=progress_bar):
     filepath = os.path.join(working_directory, filename)
 
     if not os.path.exists(filepath):
+
         _download(filename, working_directory, url_source)
         statinfo = os.stat(filepath)
         logging.info('Succesfully downloaded %s %s bytes.' % (filename, statinfo.st_size))  # , 'bytes.')

tensorlayer/models/vgg19.py

@@ -54,11 +54,31 @@ def vgg19_simple_api(net_in, end_with):
             # with pixels values in the range of 0-255 and subtracts the mean image
             # values (calculated over the entire ImageNet training set).
 
-            # Rescale the input tensor with pixels values in the range of 0-255
-            net_in.outputs = net_in.outputs * 255.0
-
-            mean = tf.constant([103.939, 116.779, 123.68], dtype=tf.float32, shape=[1, 1, 1, 3], name='img_mean')
-            net_in.outputs = net_in.outputs - mean
+            # # Rescale the input tensor with pixels values in the range of 0-255
+            # net_in.outputs = net_in.outputs * 255.0
+            # red, green, blue = tf.split(net_in.outputs, 3, 3)
+            VGG_MEAN = [103.939, 116.779, 123.68]
+            #
+            # bgr = tf.concat([
+            #     blue - VGG_MEAN[0],
+            #     green - VGG_MEAN[1],
+            #     red - VGG_MEAN[2],
+            # ], axis=3)
+            # # mean = tf.constant([103.939, 116.779, 123.68], dtype=tf.float32, shape=[1, 1, 1, 3], name='img_mean')
+            # # net_in.outputs = net_in.outputs - mean
+            # net_in.outputs = bgr
+            rgb = net_in.outputs
+            rgb_scaled = rgb * 255.0
+            # Convert RGB to BGR
+            red, green, blue = tf.split(rgb_scaled, 3, 3)
+
+            bgr = tf.concat([
+                blue - VGG_MEAN[0],
+                green - VGG_MEAN[1],
+                red - VGG_MEAN[2],
+            ], axis=3)
+
+            net_in.outputs = bgr
 
         layers = [
             # conv1
@@ -163,6 +183,8 @@ def restore_params(self, sess):
             b = np.asarray(val[1][1])
             print("  Loading %s: %s, %s" % (val[0], W.shape, b.shape))
             params.extend([W, b])
+            if len(self.all_params) == len(params):
+                break
 
         print("Restoring model from npz file")
         assign_params(sess, params, self.net)