small changes; fix travis

.travis.yml

@@ -26,10 +26,10 @@ matrix:
     env: TF_VERSION=1.3.0 TF_TYPE=release
   - os: linux
     python: 2.7
-    env: TF_VERSION=1.10.0 TF_TYPE=release
+    env: TF_VERSION=1.12.0 TF_TYPE=release
   - os: linux
     python: 3.6
-    env: TF_VERSION=1.10.0 TF_TYPE=release PYPI=true
+    env: TF_VERSION=1.12.0 TF_TYPE=release PYPI=true
   - os: linux
     python: 2.7
     env: TF_TYPE=nightly
@@ -43,7 +43,7 @@ install:
 - pip install -U pip    # the pip version on travis is too old
 - pip install .
 - pip install flake8 scikit-image opencv-python lmdb h5py msgpack
-  # check that dataflow can be imported alone
+  # check that dataflow can be imported alone without tensorflow
 - python -c "import tensorpack.dataflow"
 - ./tests/install-tensorflow.sh
 

examples/FasterRCNN/model_frcnn.py

@@ -222,9 +222,9 @@ def fastrcnn_predictions(boxes, scores):
         return mask
 
     # TF bug in version 1.11, 1.12: https://github.com/tensorflow/tensorflow/issues/22750
-    parallel = 1 if (get_tf_version_tuple() in [(1, 11), (1, 12)]) else 10
+    buggy_tf = get_tf_version_tuple() in [(1, 11), (1, 12)]
     masks = tf.map_fn(f, (scores, boxes), dtype=tf.bool,
-                      parallel_iterations=parallel)     # #cat x N
+                      parallel_iterations=1 if buggy_tf else 10)     # #cat x N
     selected_indices = tf.where(masks)  # #selection x 2, each is (cat_id, box_id)
     scores = tf.boolean_mask(scores, masks)
 

tensorpack/dataflow/imgaug/__init__.py

@@ -16,6 +16,7 @@ if STATICA_HACK:
     from .noise import *
     from .paste import *
     from .transform import *
+    from .external import *
 
 
 import os

tensorpack/dataflow/imgaug/external.py

@@ -33,12 +33,12 @@ class IAAugmentor(ImageAugmentor):
     def _get_augment_params(self, img):
         return (self._aug.to_deterministic(), img.shape)
 
-    def _augment(self, img, p):
-        aug, _ = p
+    def _augment(self, img, param):
+        aug, _ = param
         return aug.augment_image(img)
 
-    def _augment_coords(self, coords, p):
-        aug, shape = p
+    def _augment_coords(self, coords, param):
+        aug, shape = param
         points = [IA.Keypoint(x=x, y=y) for x, y in coords]
         points = IA.KeypointsOnImage(points, shape=shape)
         augmented = aug.augment_keypoints([points])[0].keypoints

tensorpack/libinfo.py

@@ -6,6 +6,7 @@ os.environ['OPENCV_OPENCL_RUNTIME'] = 'disabled'     # https://github.com/opencv
 try:
     # issue#1924 may happen on old systems
     import cv2  # noqa
+    # cv2.setNumThreads(0)
     if int(cv2.__version__.split('.')[0]) == 3:
         cv2.ocl.setUseOpenCL(False)
     # check if cv is built with cuda or openmp