Merge remote-tracking branch 'flip/master'

datasets/imdb.py

 import os
+import PIL
 
 class imdb(object):
     def __init__(self, name):
@@ -63,3 +64,22 @@ class imdb(object):
         all_boxes[class][image] = [] or np.array of shape #dets x 5
         """
         raise NotImplementedError
+
+    def append_flipped_roidb(self):
+        num_images = len(self.image_index)
+        widths = [PIL.Image.open(self.image_path_at(i)).size[0]
+                  for i in xrange(num_images)]
+        for i in xrange(num_images):
+            self.roidb[i]['flipped'] = False
+            boxes = self.roidb[i]['boxes'].copy()
+            oldx1 = boxes[:, 0].copy()
+            oldx2 = boxes[:, 2].copy()
+            boxes[:, 0] = widths[i] - oldx2 - 1
+            boxes[:, 2] = widths[i] - oldx1 - 1
+            assert (boxes[:, 2] >= boxes[:, 0]).all()
+            entry = {'boxes' : boxes,
+                     'gt_overlaps' : self.roidb[i]['gt_overlaps'],
+                     'gt_classes' : self.roidb[i]['gt_classes'],
+                     'flipped' : True}
+            self.roidb.append(entry)
+        self._image_index = self._image_index * 2

datasets/pascal_voc.py

@@ -58,7 +58,7 @@ class pascal_voc(datasets.imdb):
                                       self._image_set + '.txt')
         assert os.path.exists(image_set_file)
         with open(image_set_file) as f:
-            image_index = tuple([x.strip() for x in f.readlines()])
+            image_index = [x.strip() for x in f.readlines()]
         return image_index
 
     def _get_default_path(self):

fast_rcnn_train.py

@@ -156,6 +156,9 @@ if __name__ == '__main__':
 
     imdb_train = datasets.pascal_voc('trainval', '2007')
 
+    # enhance roidb to contain flipped examples
+    imdb_train.append_flipped_roidb()
+
     # enhance roidb to contain some useful derived quanties
     roidb_train = training_roidb(imdb_train)
 

finetuning.py

@@ -129,6 +129,8 @@ def _get_image_blob(roidb, scale_inds):
     im_scale_factors = []
     for i in xrange(num_images):
         im = cv2.imread(roidb[i]['image'])
+        if roidb[i]['flipped']:
+            im = im[:, ::-1, :]
         im = im.astype(np.float32, copy=False)
         im -= conf.PIXEL_MEANS
         im_shape = im.shape