Merge pull request #2952 from activeloopai/upgrade_pillow

Upgrade pillow

deeplake/core/chunk/sample_compressed_chunk.py

@@ -158,11 +158,11 @@ def read_sample( # type: ignore
  return sample
 
  if squeeze:
- sample = sample.squeeze(0)
+ sample = sample.squeeze(0) # type: ignore
 
- if cast and sample.dtype != self.dtype:
- sample = sample.astype(self.dtype)
- elif copy and not sample.flags["WRITEABLE"]:
+ if cast and sample.dtype != self.dtype: # type: ignore
+ sample = sample.astype(self.dtype) # type: ignore
+ elif copy and not sample.flags["WRITEABLE"]: # type: ignore
  sample = sample.copy()
  return sample
 

deeplake/core/compression.py

@@ -86,7 +86,7 @@
 _STRUCT_II = struct.Struct(">ii")
 
 
-def to_image(array: np.ndarray) -> Image:
+def to_image(array: np.ndarray) -> Image.Image:
  shape = array.shape
  if len(shape) == 3 and shape[0] != 1 and shape[2] == 1:
  # convert (X,Y,1) grayscale to (X,Y) for pillow compatibility
@@ -116,17 +116,22 @@ def _compress_apng(array: np.ndarray) -> bytes:
 
 def _decompress_apng(buffer: Union[bytes, memoryview]) -> np.ndarray:
  img = Image.open(BytesIO(buffer))
+
+ n_frames = getattr(img, "n_frames", 1)
+
+ if n_frames == 1:
+ raise ValueError("Image does not support multiple frames.")
  frame0 = np.array(img)
  if frame0.ndim == 2:
- ret = np.zeros(frame0.shape + (img.n_frames,), dtype=frame0.dtype)
+ ret = np.zeros(frame0.shape + (n_frames,), dtype=frame0.dtype)
  ret[:, :, 0] = frame0
- for i in range(1, img.n_frames):
+ for i in range(1, n_frames):
  img.seek(i)
  ret[:, :, i] = np.array(img)
  else:
- ret = np.zeros((img.n_frames,) + frame0.shape, dtype=frame0.dtype)
+ ret = np.zeros((n_frames,) + frame0.shape, dtype=frame0.dtype)
  ret[0] = frame0
- for i in range(1, img.n_frames):
+ for i in range(1, n_frames):
  img.seek(i)
  ret[i] = np.array(img)
  return ret
@@ -420,9 +425,9 @@ def decompress_multiple(
  next_x = 0
  for shape in shapes:
  if shape == (0, 0, 0):
- arrays.append(np.zeros(shape, dtype=canvas.dtype))
+ arrays.append(np.zeros(shape, dtype=canvas.dtype)) # type: ignore
  else:
- arrays.append(canvas[: shape[0], next_x : next_x + shape[1]])
+ arrays.append(canvas[: shape[0], next_x : next_x + shape[1]]) # type: ignore
  next_x += shape[1]
  return arrays
 
@@ -728,7 +733,7 @@ def read_meta_from_compressed_file(
  else:
  img = Image.open(f) if isfile else Image.open(BytesIO(f)) # type: ignore
  shape, typestr = Image._conv_type_shape(img)
- compression = img.format.lower()
+ compression = None if img.format is None else img.format.lower()
  return compression, shape, typestr # type: ignore
  finally:
  if close:

deeplake/core/sample.py

@@ -392,10 +392,10 @@ def _decompress(self, to_pil: bool = False):
  compression=compression,
  shape=self.shape,
  dtype=self.dtype,
- )
- self._uncompressed_bytes = self._array.tobytes()
- self._typestr = self._array.__array_interface__["typestr"]
- self._dtype = np.dtype(self._typestr).name
+ ) # type: ignore
+ self._uncompressed_bytes = self._array.tobytes() # type: ignore
+ self._typestr = self._array.__array_interface__["typestr"] # type: ignore
+ self._dtype = np.dtype(self._typestr).name # type: ignore
 
  def uncompressed_bytes(self) -> Optional[bytes]:
  """Returns uncompressed bytes."""
@@ -420,7 +420,7 @@ def array(self) -> np.ndarray: # type: ignore
  return self._array # type: ignore
 
  @property
- def pil(self) -> Image.Image: # type: ignore
+ def pil(self) -> Optional[Image.Image]: # type: ignore
  """Return PIL image corresponding to the sample. Decompresses the sample if necessary.
 
  Example:

deeplake/requirements/common.txt

@@ -1,5 +1,5 @@
 numpy<2.0
-pillow~=10.2.0
+pillow~=10.4.0
 boto3
 click
 google-cloud-storage~=1.42.0

deeplake/requirements/plugins.txt

@@ -1,11 +1,11 @@
---find-links https://download.openmmlab.com/mmcv/dist/cpu/torch1.13/index.html
+--find-links https://download.openmmlab.com/mmcv/dist/cpu/torch1.12/index.html
 torch
 torchvision
 tensorflow
 tensorflow_datasets
 pickle5>=0.0.11; python_version < "3.8" and python_version >= "3.6"
 datasets~=1.17
-mmcv-full==1.7.1; platform_system == "Linux" and python_version >= "3.7"
+mmcv-full==1.7.1; platform_system == "Linux" and python_version >= "3.7" --find-links https://download.openmmlab.com/mmcv/dist/cpu/torch1.12/index.html
 mmdet==2.28.1; platform_system == "Linux" and python_version >= "3.7"
 mmsegmentation==0.30.0; platform_system == "Linux" and python_version >= "3.7"
 mmengine

deeplake/tests/common.py

@@ -78,7 +78,7 @@ def get_actual_compression_from_buffer(buffer: memoryview) -> Optional[str]:
  try:
  bio = BytesIO(buffer)
  img = Image.open(bio)
- return img.format.lower()
+ return None if img.format is None else img.format.lower()
 
  except UnidentifiedImageError:
  return None

deeplake/util/downsample.py

@@ -30,7 +30,7 @@ def needs_downsampling(sample, factor: int):
  if isinstance(sample, Image.Image):
  dimensions = sample.size
  elif isinstance(sample, np.ndarray):
- dimensions = sample.shape
+ dimensions = sample.shape # type: ignore
  if len(dimensions) == 3 and dimensions[2] == 0:
  return False
  dimensions = dimensions[:2]
@@ -62,7 +62,7 @@ def downsample_sample(
  elif isinstance(sample, SampleTiles):
  sample = sample.arr or PartialSample(
  sample.sample_shape, sample.tile_shape, sample.dtype
- )
+ ) # type: ignore
 
  if isinstance(sample, PartialSample):
  return sample.downsample(factor)
@@ -126,7 +126,7 @@ def downsample_link_tiled(
  if compression is None:
  return arr
  with io.BytesIO() as f:
- Image.fromarray(arr).save(f, format=compression)
+ Image.fromarray(arr).save(f, format=compression) # type: ignore
  image_bytes = f.getvalue()
  return deeplake.core.sample.Sample(buffer=image_bytes, compression=compression)
 

deeplake/util/exif.py

@@ -1,4 +1,5 @@
 from fractions import Fraction
+from typing import Any, Dict
 
 from PIL import Image # type: ignore
 import PIL.ExifTags # type: ignore
@@ -41,22 +42,23 @@
 }
 
 
-def getexif(image: Image):
- raw_exif = image._getexif()
+def getexif(image: Image.Image) -> Dict[str, Any]:
+ raw_exif = image.getexif()
  if not raw_exif:
  return {}
+
  exif = {}
  for k, v in raw_exif.items():
  tag = k
- k = PIL.ExifTags.TAGS.get(k, k)
- v = _process_exif_value(k, v)
+ key = str(PIL.ExifTags.TAGS.get(k, k))
+ v = _process_exif_value(key, v)
  if v is None:
  continue
- exif[k] = v
+ exif[key] = v
  return exif
 
 
-def _process_exif_value(k: str, v):
+def _process_exif_value(k: str, v: Any) -> Any:
  if hasattr(v, "numerator"): # Rational
  v = v.numerator / v.denominator if v.denominator else 0
  if k == "ExposureTime":
@@ -67,12 +69,18 @@ def _process_exif_value(k: str, v):
  except ValueError: # nan
  pass
  return v
+
  elif k in _LOOKUPS:
- return _LOOKUPS[k][v]
- elif isinstance(v, bytes):
+ if isinstance(_LOOKUPS[k], dict):
+ return _LOOKUPS[k].get(v, v) # type: ignore
+ elif isinstance(_LOOKUPS[k], tuple):
+ return _LOOKUPS[k][v] if v < len(_LOOKUPS[k]) else v
+
+ elif isinstance(v, bytes): # Handle byte data
  return str(v) if len(v) < 16 else "<bytes>"
  elif isinstance(v, (list, tuple)):
  return type(v)(_process_exif_value("_", x) for x in v)
  elif isinstance(v, dict):
- return {k: _process_exif_value("_", x) for k, x in v.items()}
+ return {key: _process_exif_value("_", x) for key, x in v.items()}
+
  return str(v)

deeplake/util/image.py

@@ -13,7 +13,7 @@ def convert_sample(image_sample: Sample, mode: str) -> Sample:
  image.close()
  return image_sample
 
- image = image.convert(mode)
+ image = image.convert(mode) # type: ignore
  image_bytes = BytesIO()
  image.save(image_bytes, format=image_sample.compression)
  converted = Sample(