EXIF orientation support (#4529)

Co-authored-by: Andrey Zhavoronkov <andrey.zhavoronkov@intel.com>
Co-authored-by: Nikita Manovich <nikita.manovich@intel.com>
Co-authored-by: Maria Khrustaleva <maya17grd@gmail.com>

CHANGELOG.md

@@ -30,6 +30,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - some AI Tools were not sending responses properly (<https://github.com/openvinotoolkit/cvat/issues/4432>)
 - Unable to upload annotations (<https://github.com/openvinotoolkit/cvat/pull/4513>)
 - Fix build dependencies for Siammask (<https://github.com/openvinotoolkit/cvat/pull/4486>)
+- Bug: Exif orientation information handled incorrectly (<https://github.com/openvinotoolkit/cvat/pull/4529>)
 
 ### Security
 - TDB

cvat-core/src/frames.js

@@ -66,6 +66,14 @@
                         value: height,
                         writable: false,
                     },
+                    /**
+                     * task ID
+                     * @name tid
+                     * @type {integer}
+                     * @memberof module:API.cvat.classes.FrameData
+                     * @readonly
+                     * @instance
+                     */
                     tid: {
                         value: taskID,
                         writable: false,

cvat/apps/engine/media_extractors.py

@@ -9,6 +9,7 @@ import zipfile
 import io
 import itertools
 import struct
+from enum import IntEnum
 from abc import ABC, abstractmethod
 from contextlib import closing
 
@@ -29,6 +30,20 @@ ImageFile.LOAD_TRUNCATED_IMAGES = True
 from cvat.apps.engine.mime_types import mimetypes
 from utils.dataset_manifest import VideoManifestManager, ImageManifestManager
 
+ORIENTATION_EXIF_TAG = 274
+
+
+class ORIENTATION(IntEnum):
+    NORMAL_HORIZONTAL=1
+    MIRROR_HORIZONTAL=2
+    NORMAL_180_ROTATED=3
+    MIRROR_VERTICAL=4
+    MIRROR_HORIZONTAL_270_ROTATED=5
+    NORMAL_90_ROTATED=6
+    MIRROR_HORIZONTAL_90_ROTATED=7
+    NORMAL_270_ROTATED=8
+
+
 def get_mime(name):
     for type_name, type_def in MEDIA_TYPES.items():
         if type_def['has_mime_type'](name):
@@ -62,6 +77,27 @@ def sort(images, sorting_method=SortingMethod.LEXICOGRAPHICAL, func=None):
     else:
         raise NotImplementedError()
 
+def image_size_within_orientation(img: Image):
+    orientation = img.getexif().get(ORIENTATION_EXIF_TAG, ORIENTATION.NORMAL_HORIZONTAL)
+    if orientation > 4:
+        return img.height, img.width
+    return img.width, img.height
+
+def rotate_within_exif(img: Image):
+    orientation = img.getexif().get(ORIENTATION_EXIF_TAG,  ORIENTATION.NORMAL_HORIZONTAL)
+    if orientation in [ORIENTATION.NORMAL_180_ROTATED, ORIENTATION.MIRROR_VERTICAL]:
+        img = img.rotate(180, expand=True)
+    elif orientation in [ORIENTATION.NORMAL_270_ROTATED, ORIENTATION.MIRROR_HORIZONTAL_90_ROTATED]:
+        img = img.rotate(90, expand=True)
+    elif orientation in [ORIENTATION.NORMAL_90_ROTATED, ORIENTATION.MIRROR_HORIZONTAL_270_ROTATED]:
+        img = img.rotate(270, expand=True)
+    if orientation in [
+        ORIENTATION.MIRROR_HORIZONTAL, ORIENTATION.MIRROR_VERTICAL,
+        ORIENTATION.MIRROR_HORIZONTAL_270_ROTATED ,ORIENTATION.MIRROR_HORIZONTAL_90_ROTATED,
+    ]:
+        img = img.transpose(Image.FLIP_LEFT_RIGHT)
+    return img
+
 class IMediaReader(ABC):
     def __init__(self, source_path, step, start, stop, dimension):
         self._source_path = source_path
@@ -85,11 +121,13 @@ class IMediaReader(ABC):
     @staticmethod
     def _get_preview(obj):
         PREVIEW_SIZE = (256, 256)
+
         if isinstance(obj, io.IOBase):
             preview = Image.open(obj)
         else:
             preview = obj
         preview.thumbnail(PREVIEW_SIZE)
+        preview = rotate_within_exif(preview)
 
         return preview.convert('RGB')
 
@@ -173,7 +211,7 @@ class ImageListReader(IMediaReader):
                 properties = ValidateDimension.get_pcd_properties(f)
                 return int(properties["WIDTH"]),  int(properties["HEIGHT"])
         img = Image.open(self._source_path[i])
-        return img.width, img.height
+        return image_size_within_orientation(img)
 
     def reconcile(self, source_files, step=1, start=0, stop=None, dimension=DimensionType.DIM_2D, sorting_method=None):
         # FIXME
@@ -314,7 +352,7 @@ class ZipReader(ImageListReader):
                 properties = ValidateDimension.get_pcd_properties(f)
                 return int(properties["WIDTH"]),  int(properties["HEIGHT"])
         img = Image.open(io.BytesIO(self._zip_source.read(self._source_path[i])))
-        return img.width, img.height
+        return image_size_within_orientation(img)
 
     def get_image(self, i):
         if self._dimension == DimensionType.DIM_3D:
@@ -538,6 +576,7 @@ class IChunkWriter(ABC):
     @staticmethod
     def _compress_image(image_path, quality):
         image = image_path.to_image() if isinstance(image_path, av.VideoFrame) else Image.open(image_path)
+        image = rotate_within_exif(image)
         # Ensure image data fits into 8bit per pixel before RGB conversion as PIL clips values on conversion
         if image.mode == "I":
             # Image mode is 32bit integer pixels.

utils/dataset_manifest/core.py

@@ -194,14 +194,18 @@ class DatasetImagesReader:
             if idx in self.range_:
                 image = next(sources)
                 img = Image.open(image, mode='r')
+                orientation = img.getexif().get(274, 1)
                 img_name = os.path.relpath(image, self._data_dir) if self._data_dir \
                     else os.path.basename(image)
                 name, extension = os.path.splitext(img_name)
+                width, height = img.width, img.height
+                if orientation > 4:
+                    width, height = height, width
                 image_properties = {
                     'name': name.replace('\\', '/'),
                     'extension': extension,
-                    'width': img.width,
+                    'width': width,
-                    'height': img.height,
+                    'height': height,
                 }
                 if self._meta and img_name in self._meta:
                     image_properties['meta'] = self._meta[img_name]