def test_decode_png(img_path, pil_mode, mode):
with Image.open(img_path) as img:
if pil_mode is not None:
img = img.convert(pil_mode)
img_pil = torch.from_numpy(np.array(img))
img_pil = normalize_dimensions(img_pil)
if img_path.endswith("16.png"):
# 16 bits image decoding is supported, but only as a private API
# FIXME: see https://github.com/pytorch/vision/issues/4731 for potential solutions to making it public
with pytest.raises(RuntimeError,
match="At most 8-bit PNG images are supported"):
data = read_file(img_path)
img_lpng = decode_image(data, mode=mode)
img_lpng = _read_png_16(img_path, mode=mode)
assert img_lpng.dtype == torch.int32
# PIL converts 16 bits pngs in uint8
img_lpng = torch.round(img_lpng / (2**16 - 1) * 255).to(torch.uint8)
else:
data = read_file(img_path)
img_lpng = decode_image(data, mode=mode)
tol = 0 if pil_mode is None else 1
if PILLOW_VERSION >= (8, 3) and pil_mode == "LA":
# Avoid checking the transparency channel until
# https://github.com/python-pillow/Pillow/issues/5593#issuecomment-878244910
# is fixed.
# TODO: remove once fix is released in PIL. Should be > 8.3.1.
img_lpng, img_pil = img_lpng[0], img_pil[0]
torch.testing.assert_close(img_lpng, img_pil, atol=tol, rtol=0)
def test_decode_image(self):
for img_path in get_images(IMAGE_ROOT, ".jpg"):
img_pil = torch.load(img_path.replace('jpg', 'pth'))
img_pil = img_pil.permute(2, 0, 1)
img_ljpeg = decode_image(read_file(img_path))
self.assertTrue(img_ljpeg.equal(img_pil))
for img_path in get_images(IMAGE_DIR, ".png"):
img_pil = torch.from_numpy(np.array(Image.open(img_path)))
img_pil = img_pil.permute(2, 0, 1)
img_lpng = decode_image(read_file(img_path))
self.assertTrue(img_lpng.equal(img_pil))
def test_decode_jpeg(self):
conversion = [(None, 0), ("L", 1), ("RGB", 3)]
for img_path in get_images(IMAGE_ROOT, ".jpg"):
for pil_mode, channels in conversion:
with Image.open(img_path) as img:
is_cmyk = img.mode == "CMYK"
if pil_mode is not None:
if is_cmyk:
# libjpeg does not support the conversion
continue
img = img.convert(pil_mode)
img_pil = torch.from_numpy(np.array(img))
if is_cmyk:
# flip the colors to match libjpeg
img_pil = 255 - img_pil
img_pil = normalize_dimensions(img_pil)
data = read_file(img_path)
img_ljpeg = decode_image(data, channels=channels)
# Permit a small variation on pixel values to account for implementation
# differences between Pillow and LibJPEG.
abs_mean_diff = (img_ljpeg.type(torch.float32) - img_pil).abs().mean().item()
self.assertTrue(abs_mean_diff < 2)
with self.assertRaisesRegex(RuntimeError, "Expected a non empty 1-dimensional tensor"):
decode_jpeg(torch.empty((100, 1), dtype=torch.uint8))
with self.assertRaisesRegex(RuntimeError, "Expected a torch.uint8 tensor"):
decode_jpeg(torch.empty((100, ), dtype=torch.float16))
with self.assertRaises(RuntimeError):
decode_jpeg(torch.empty((100), dtype=torch.uint8))
def test_decode_jpeg_cuda(mode, img_path, scripted):
if "cmyk" in img_path:
pytest.xfail("Decoding a CMYK jpeg isn't supported")
data = read_file(img_path)
img = decode_image(data, mode=mode)
f = torch.jit.script(decode_jpeg) if scripted else decode_jpeg
img_nvjpeg = f(data, mode=mode, device="cuda")
# Some difference expected between jpeg implementations
assert (img.float() - img_nvjpeg.cpu().float()).abs().mean() < 2
def test_decode_png(img_path, pil_mode, mode):
with Image.open(img_path) as img:
if pil_mode is not None:
img = img.convert(pil_mode)
img_pil = torch.from_numpy(np.array(img))
img_pil = normalize_dimensions(img_pil)
data = read_file(img_path)
img_lpng = decode_image(data, mode=mode)
tol = 0 if pil_mode is None else 1
assert img_lpng.allclose(img_pil, atol=tol)
def test_decode_png(self):
conversion = [(None, 0), ("L", 1), ("LA", 2), ("RGB", 3), ("RGBA", 4)]
for img_path in get_images(FAKEDATA_DIR, ".png"):
for pil_mode, channels in conversion:
with Image.open(img_path) as img:
if pil_mode is not None:
img = img.convert(pil_mode)
img_pil = torch.from_numpy(np.array(img))
img_pil = normalize_dimensions(img_pil)
data = read_file(img_path)
img_lpng = decode_image(data, channels=channels)
tol = 0 if conversion is None else 1
self.assertTrue(img_lpng.allclose(img_pil, atol=tol))
with self.assertRaises(RuntimeError):
decode_png(torch.empty((), dtype=torch.uint8))
with self.assertRaises(RuntimeError):
decode_png(torch.randint(3, 5, (300,), dtype=torch.uint8))
def test_decode_png(img_path, pil_mode, mode):
with Image.open(img_path) as img:
if pil_mode is not None:
img = img.convert(pil_mode)
img_pil = torch.from_numpy(np.array(img))
img_pil = normalize_dimensions(img_pil)
data = read_file(img_path)
img_lpng = decode_image(data, mode=mode)
tol = 0 if pil_mode is None else 1
if PILLOW_VERSION >= (8, 3) and pil_mode == "LA":
# Avoid checking the transparency channel until
# https://github.com/python-pillow/Pillow/issues/5593#issuecomment-878244910
# is fixed.
# TODO: remove once fix is released in PIL. Should be > 8.3.1.
img_lpng, img_pil = img_lpng[0], img_pil[0]
torch.testing.assert_close(img_lpng, img_pil, atol=tol, rtol=0)
def test_decode_jpeg(img_path, pil_mode, mode):
with Image.open(img_path) as img:
is_cmyk = img.mode == "CMYK"
if pil_mode is not None:
if is_cmyk:
# libjpeg does not support the conversion
pytest.xfail("Decoding a CMYK jpeg isn't supported")
img = img.convert(pil_mode)
img_pil = torch.from_numpy(np.array(img))
if is_cmyk:
# flip the colors to match libjpeg
img_pil = 255 - img_pil
img_pil = normalize_dimensions(img_pil)
data = read_file(img_path)
img_ljpeg = decode_image(data, mode=mode)
# Permit a small variation on pixel values to account for implementation
# differences between Pillow and LibJPEG.
abs_mean_diff = (img_ljpeg.type(torch.float32) - img_pil).abs().mean().item()
assert abs_mean_diff < 2
