Fix exif image orientation (facebookresearch#262)

Summary:
hi.

I tried demo.py our data.
I found small problem for image have exif information.

I fixed it. please check it. thanks.
Pull Request resolved: facebookresearch#262

Differential Revision: D18424793

Pulled By: ppwwyyxx

fbshipit-source-id: 3660f7d8637027b6aafe51801fe1f716bc55b933

detectron2/data/detection_utils.py

@@ -9,7 +9,7 @@
 import numpy as np
 import torch
 from fvcore.common.file_io import PathManager
-from PIL import Image
+from PIL import Image, ImageOps
 
 from detectron2.structures import (
     BitMasks,
@@ -34,6 +34,7 @@ class SizeMismatchError(ValueError):
 def read_image(file_name, format=None):
     """
     Read an image into the given format.
+    Will apply rotation and flipping if the image has such exif information.
 
     Args:
         file_name (str): image file path
@@ -45,6 +46,8 @@ def read_image(file_name, format=None):
     with PathManager.open(file_name, "rb") as f:
         image = Image.open(f)
 
+        image = ImageOps.exif_transpose(image)
+
         if format is not None:
             # PIL only supports RGB, so convert to RGB and flip channels over below
             conversion_format = format

docs/notes/changelog.md

@@ -6,6 +6,10 @@
 * 2019-10-10: initial release.
 
 
+### Minor Changes in Code
+
+* 2019-11-11: detectron2.data.detection_utils.read_image is transposing with exif information.
+
 ### Config Version Change Log
 
 * v1: Rename `RPN_HEAD.NAME` to `RPN.HEAD_NAME`.

docs/tutorials/datasets.md

@@ -42,7 +42,7 @@ The fields are often optional, and some functions may be able to
 infer certain fields from others if needed, e.g., the data loader
 can load an image from "file_name" if the "image" field is not available.
 
-+ `file_name`: the full path to the image file.
++ `file_name`: the full path to the image file. Will apply rotation and flipping if the image has such exif information.
 + `sem_seg_file_name`: the full path to the ground truth semantic segmentation file.
 + `image`: the image as a numpy array.
 + `sem_seg`: semantic segmentation ground truth in a 2D numpy array. Values in the array represent
@@ -160,3 +160,4 @@ NOTE: For background on the difference between "thing" and "stuff" categories, s
 In detectron2, the term "thing" is used for instance-level tasks,
 and "stuff" is used for semantic segmentation tasks.
 Both are used in panoptic segmentation.
+

setup.py

@@ -72,7 +72,7 @@ def get_extensions():
     python_requires=">=3.6",
     install_requires=[
         "termcolor>=1.1",
-        "Pillow",
+        "Pillow>=6.0",
         "yacs>=0.1.6",
         "tabulate",
         "cloudpickle",