Skip to content

Commit

Permalink
Fix face image parsing bugs
Browse files Browse the repository at this point in the history
  • Loading branch information
@derneuere
derneuere committed Dec 25, 2023
1 parent c4e5871 commit 55d8f49
Show file tree
Hide file tree
Showing 3 changed files with 47 additions and 54 deletions.
1 change: 1 addition & 0 deletions api/exif_tags.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -21,3 +21,4 @@ class Tags:
SUBJECT_DISTANCE = "EXIF:SubjectDistance"
DIGITAL_ZOOM_RATIO = "EXIF:DigitalZoomRatio"
REGION_INFO = "XMP:RegionInfo"
ROTATION = "QuickTime:Rotation"
92 changes: 38 additions & 54 deletions api/models/photo.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -30,7 +30,7 @@
doesStaticThumbnailExists,
doesVideoThumbnailExists,
)
from api.util import get_metadata, logger
from api.util import get_metadata, is_number, logger


class VisiblePhotoManager(models.Manager):
Expand Down Expand Up @@ -596,13 +596,17 @@ def _extract_faces(self, second_try=False):
api.models.cluster.get_unknown_cluster(user=self.owner)
)

(region_info,) = get_metadata(
self.main_file.path, tags=[Tags.REGION_INFO], try_sidecar=True, struct=True
(region_info, rotation) = get_metadata(
self.main_file.path,
tags=[Tags.REGION_INFO, Tags.ROTATION],
try_sidecar=True,
struct=True,
)

if region_info:
logger.debug(f"Extracted region_info for {self.main_file.path}")
logger.debug(f"region_info: {region_info}")
logger.info(f"orientation: {rotation}")
# Extract faces
for region in region_info["RegionList"]:
if region.get("Type") != "Face":
Expand All @@ -617,75 +621,55 @@ def _extract_faces(self, second_try=False):
else:
person = api.models.person.get_unknown_person(owner=self.owner)
# Create face from the region infos
image = np.array(PIL.Image.open(self.thumbnail_big.path))
big_thumbnail_image = np.array(PIL.Image.open(self.thumbnail_big.path))
area = region.get("Area")
applied_to_dimensions = region.get("AppliedToDimensions")
if (area and area.get("Unit") == "normalized") or (
applied_to_dimensions
and applied_to_dimensions.get("Unit") == "pixel"
):
# To-Do: Not sure, when to use this instead of the thumbnail size tbh
# if applied_to_dimensions:
# image_width = applied_to_dimensions.get("W")
# image_height = applied_to_dimensions.get("H")

# To-Do: Rotation, this is already handled by thumbnail creation?!
# if region.get("Rotation"):
# rotation = region.get("Rotation")
# if rotation == 90:
# image = np.rot90(image, 1)
# elif rotation == 180:
# image = np.rot90(image, 2)
# elif rotation == 270:
# image = np.rot90(image, 3)
image_width = image.shape[1]
image_height = image.shape[0]
if not area.get("X") or not isinstance(
area.get("X"), numbers.Number
):
logger.info(
f"Broken face area exif data! region_info: {region_info}"
)
continue
if not area.get("Y") or not isinstance(
area.get("Y"), numbers.Number
image_width = big_thumbnail_image.shape[1]
image_height = big_thumbnail_image.shape[0]
if (
not is_number(area.get("X"))
or not is_number(area.get("Y"))
or not is_number(area.get("W"))
or not is_number(area.get("H"))
):
logger.info(
f"Broken face area exif data! region_info: {region_info}"
f"Broken face area exif data! No numerical positional data. region_info: {region_info}"
)
continue
if not area.get("W") or not isinstance(
area.get("W"), numbers.Number
):
logger.info(
f"Broken face area exif data! region_info: {region_info}"
)
continue
if not area.get("H") or not isinstance(
area.get("H"), numbers.Number
):
logger.info(
f"Broken face area exif data! region_info: {region_info}"
)
continue

correct_w = float(area["W"])
correct_h = float(area["H"])

correct_x = float(area["X"])
correct_y = float(area["Y"])
correct_w = float(area.get("W"))
correct_h = float(area.get("H"))
correct_x = float(area.get("X"))
correct_y = float(area.get("Y"))
if rotation == 90:
temp_x = correct_x
correct_x = correct_y
correct_y = 1 - temp_x
correct_w, correct_h = correct_h, correct_w
elif rotation == 180:
correct_x = 1 - correct_x
correct_y = 1 - correct_y
elif rotation == 270:
temp_x = correct_x
correct_x = 1 - correct_y
correct_y = temp_x
correct_w, correct_h = correct_h, correct_w

# Calculate the half-width and half-height of the box
half_width = correct_w * image_width / 2
half_height = correct_h * image_height / 2
half_width = (correct_w * image_width) / 2
half_height = (correct_h * image_height) / 2

# Calculate the top, right, bottom, and left coordinates
top = int((correct_y * image_height) - half_height)
right = int((correct_x * image_width) + half_width)
bottom = int((correct_y * image_height) + half_height)
left = int((correct_x * image_width) - half_width)

face_image = image[top:bottom, left:right]
face_image = big_thumbnail_image[top:bottom, left:right]
face_image = PIL.Image.fromarray(face_image)

# Figure out which face idx it is, but reading the number of the faces of the person
Expand Down Expand Up @@ -720,7 +704,7 @@ def _extract_faces(self, second_try=False):
return

try:
image = np.array(PIL.Image.open(self.thumbnail_big.path))
big_thumbnail_image = np.array(PIL.Image.open(self.thumbnail_big.path))

face_locations = []
# Create
Expand All @@ -743,7 +727,7 @@ def _extract_faces(self, second_try=False):
face_encoding = face[0]
face_location = face[1]
top, right, bottom, left = face_location
face_image = image[top:bottom, left:right]
face_image = big_thumbnail_image[top:bottom, left:right]
face_image = PIL.Image.fromarray(face_image)

image_path = self.image_hash + "_" + str(idx_face) + ".jpg"
Expand Down
8 changes: 8 additions & 0 deletions api/util.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,14 @@
logger.setLevel(logging.INFO)


def is_number(s):
try:
float(s)
return True
except Exception:
return False


def convert_to_degrees(values):
"""
Helper function to convert the GPS coordinates stored in the EXIF to degrees in float format
Expand Down

0 comments on commit 55d8f49

Please # to comment.