Spaces:
Runtime error
Runtime error
File size: 9,815 Bytes
7349148 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 |
import warnings
from typing import Optional
import torch
from jaxtyping import Float
from lxml import etree
def load_asc_cdl(cdl_path: str, device: torch.device = torch.device("cpu")) -> dict:
"""
Loads ASC CDL parameters from an XML file.
Parameters:
cdl_path (str): Path to the ASC CDL XML file
Returns:
Dict:
slope, offset, power, and saturation values as torch tensors
"""
try:
tree = etree.parse(cdl_path)
root = tree.getroot()
except Exception as e:
raise ValueError(f"Error loading ASC CDL from {cdl_path}: {e}")
# Extract SOP values
sop_node = root.find(".//SOPNode")
slope = torch.tensor(
[float(x) for x in sop_node.find("Slope").text.split()], device=device
)
offset = torch.tensor(
[float(x) for x in sop_node.find("Offset").text.split()], device=device
)
power = torch.tensor(
[float(x) for x in sop_node.find("Power").text.split()], device=device
)
# Extract Saturation value
sat_node = root.find(".//SatNode")
saturation = torch.tensor(float(sat_node.find("Saturation").text), device=device)
return {"slope": slope, "offset": offset, "power": power, "saturation": saturation}
def save_asc_cdl(cdl_dict: dict, cdl_path: Optional[str]):
"""
Saves ASC CDL parameters to an XML file.
Parameters:
cdl_dict (dict): Dictionary containing slope, offset, power, and
saturation values
"""
root = etree.Element("ASC_CDL")
sop_node = etree.SubElement(root, "SOPNode")
etree.SubElement(sop_node, "Slope").text = " ".join(
str(x) for x in cdl_dict["slope"].detach().cpu().numpy()
)
etree.SubElement(sop_node, "Offset").text = " ".join(
str(x) for x in cdl_dict["offset"].detach().cpu().numpy()
)
etree.SubElement(sop_node, "Power").text = " ".join(
str(x) for x in cdl_dict["power"].detach().cpu().numpy()
)
sat_node = etree.SubElement(root, "SatNode")
etree.SubElement(sat_node, "Saturation").text = str(
cdl_dict["saturation"].detach().cpu().numpy()
)
tree = etree.ElementTree(root)
if cdl_path is not None:
try:
tree.write(
cdl_path, pretty_print=True, xml_declaration=True, encoding="utf-8"
)
except Exception as e:
raise ValueError(f"Error saving ASC CDL to {cdl_path}: {e}")
else:
return etree.tostring(
root, pretty_print=True, xml_declaration=True, encoding="utf-8"
).decode("utf-8")
def apply_sop(
img: Float[torch.Tensor, "*B C H W"],
slope: Float[torch.Tensor, "*B C"],
offset: Float[torch.Tensor, "*B C"],
power: Float[torch.Tensor, "*B C"],
clamp: bool = True,
) -> Float[torch.Tensor, "*B C H W"]:
"""
Applies Slope, Offset, and Power adjustments.
Parameters:
img (torch.Tensor): Input image tensor (*B, C, H, W)
slope (torch.Tensor): Slope per channel (*B, C)
offset (torch.Tensor): Offset per channel (*B, C)
power (torch.Tensor): Power per channel (*B, C)
Returns:
torch.Tensor: Image after SOP adjustments.
"""
so = img * slope.unsqueeze(-1).unsqueeze(-1) + offset.unsqueeze(-1).unsqueeze(-1)
if clamp:
so = torch.clamp(so, min=0.0, max=1.0)
return torch.where(
so > 1e-7, torch.pow(so.clamp(min=1e-7), power.unsqueeze(-1).unsqueeze(-1)), so
)
def apply_saturation(
img: Float[torch.Tensor, "*B C H W"],
saturation: Float[torch.Tensor, "*B"],
) -> Float[torch.Tensor, "*B C H W"]:
"""
Applies saturation adjustment.
Parameters:
img (torch.Tensor): Image tensor (*B, C, H, W)
saturation (torch.Tensor): Saturation factor (*B)
Returns:
torch.Tensor: Image after saturation adjustment.
"""
# Calculate luminance using Rec. 709 coefficients
lum = (
0.2126 * img[..., 0, :, :]
+ 0.7152 * img[..., 1, :, :]
+ 0.0722 * img[..., 2, :, :]
)
lum = lum.unsqueeze(-3) # Add channel dimension
return lum + (img - lum) * saturation.unsqueeze(-1).unsqueeze(-1).unsqueeze(-1)
def asc_cdl_forward(
img: Float[torch.Tensor, "*B C H W"],
slope: Float[torch.Tensor, "*B C"],
offset: Float[torch.Tensor, "*B C"],
power: Float[torch.Tensor, "*B C"],
saturation: Float[torch.Tensor, "*B"],
clamp: bool = True,
) -> Float[torch.Tensor, "*B C H W"]:
"""
Applies ASC CDL transformation in Fwd or FwdNoClamp mode.
Parameters:
img (torch.Tensor): Input image tensor (*B, C, H, W)
slope (torch.Tensor): Slope per channel (*B, C)
offset (torch.Tensor): Offset per channel (*B, C)
power (torch.Tensor): Power per channel (*B, C)
saturation (torch.Tensor): Saturation factor (*B)
clamp (bool): If True, clamps output to [0, 1] (Fwd mode).
If False, no clamping (FwdNoClamp mode).
Returns:
torch.Tensor: Transformed image tensor.
"""
# Add warning if saturation, slope, power are below 0
if (saturation < 0).any():
warnings.warn("Saturation is below 0, this will result in a color shift.")
if (slope < 0).any():
warnings.warn("Slope is below 0, this will result in a color shift.")
if (power < 0).any():
warnings.warn("Power is below 0, this will result in a color shift.")
img_batch_dim = img.shape[:-3]
# Check if slope, offset, power, saturation have the same batch dimension
# If they do not have any batch dimensions, add a single batch dimensions
if slope.ndim == 1:
slope = slope.view(*[1] * len(img_batch_dim), *slope.shape)
if offset.ndim == 1:
offset = offset.view(*[1] * len(img_batch_dim), *offset.shape)
if power.ndim == 1:
power = power.view(*[1] * len(img_batch_dim), *power.shape)
if saturation.ndim == 0:
saturation = saturation.view(*[1] * len(img_batch_dim), *saturation.shape)
# Now check that the lengths are matching
assert slope.ndim == len(img_batch_dim) + 1
assert offset.ndim == len(img_batch_dim) + 1
assert power.ndim == len(img_batch_dim) + 1
assert saturation.ndim == len(img_batch_dim)
# Apply Slope, Offset, and Power adjustments
img = apply_sop(img, slope, offset, power, clamp=clamp)
# print("img after sop", img.min(), img.max())
# Apply Saturation adjustment
img = apply_saturation(img, saturation)
# print("img after saturation", img.min(), img.max())
# Clamp if in Fwd mode
if clamp:
img = torch.clamp(img, 0.0, 1.0)
return img
def inverse_saturation(
img: Float[torch.Tensor, "*B C H W"],
saturation: Float[torch.Tensor, "*B"],
) -> Float[torch.Tensor, "*B C H W"]:
"""
Reverts saturation adjustment.
Parameters:
img (torch.Tensor): Image tensor (*B, C, H, W)
saturation (torch.Tensor): Saturation factor (*B)
Returns:
torch.Tensor: Image after reversing saturation adjustment.
"""
# Calculate luminance using Rec. 709 coefficients
lum = (
0.2126 * img[..., 0, :, :]
+ 0.7152 * img[..., 1, :, :]
+ 0.0722 * img[..., 2, :, :]
)
lum = lum.unsqueeze(-3) # Add channel dimension
return lum + (img - lum) / saturation.unsqueeze(-1).unsqueeze(-1).unsqueeze(-1)
def asc_cdl_reverse(
img: Float[torch.Tensor, "*B C H W"],
slope: Float[torch.Tensor, "*B C"],
offset: Float[torch.Tensor, "*B C"],
power: Float[torch.Tensor, "*B C"],
saturation: Float[torch.Tensor, "*B"],
clamp: bool = True,
) -> Float[torch.Tensor, "*B C H W"]:
"""
Applies reverse ASC CDL transformation.
Parameters:
img (torch.Tensor): Transformed image tensor (*B, C, H, W)
slope (torch.Tensor): Slope per channel (*B, C)
offset (torch.Tensor): Offset per channel (*B, C)
power (torch.Tensor): Power per channel (*B, C)
saturation (torch.Tensor): Saturation factor (*B)
clamp (bool): If True, clamps output to [0, 1].
Returns:
torch.Tensor: Recovered input image tensor.
"""
# Add warning if saturation, slope, power are below 0
if (saturation < 0).any():
warnings.warn("Saturation is below 0, this will result in a color shift.")
if (slope < 0).any():
warnings.warn("Slope is below 0, this will result in a color shift.")
if (power < 0).any():
warnings.warn("Power is below 0, this will result in a color shift.")
img_batch_dim = img.shape[:-3]
# Check if slope, offset, power, saturation have the same batch dimension
# If they do not have any batch dimensions, add a single batch dimensions
if slope.ndim == 1:
slope = slope.view(*[1] * len(img_batch_dim), *slope.shape)
if offset.ndim == 1:
offset = offset.view(*[1] * len(img_batch_dim), *offset.shape)
if power.ndim == 1:
power = power.view(*[1] * len(img_batch_dim), *power.shape)
if saturation.ndim == 0:
saturation = saturation.view(*[1] * len(img_batch_dim), *saturation.shape)
# Now check that the lengths are matching
assert slope.ndim == len(img_batch_dim) + 1
assert offset.ndim == len(img_batch_dim) + 1
assert power.ndim == len(img_batch_dim) + 1
assert saturation.ndim == len(img_batch_dim)
# Inverse Saturation adjustment
img = inverse_saturation(img, saturation)
# Inverse SOP adjustments
if clamp:
img = torch.clamp(img, 0.0, 1.0)
img = torch.where(
img > 1e-7, torch.pow(img, 1 / power.unsqueeze(-1).unsqueeze(-1)), img
)
img = (img - offset.unsqueeze(-1).unsqueeze(-1)) / slope.unsqueeze(-1).unsqueeze(-1)
# Clamp if specified
if clamp:
img = torch.clamp(img, 0.0, 1.0)
return img
|