Spaces:
Sleeping
Sleeping
File size: 6,263 Bytes
e84842d |
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 |
"""
Copyright (c) 2022, salesforce.com, inc.
All rights reserved.
SPDX-License-Identifier: BSD-3-Clause
For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause
"""
import torch
from lavis.common.registry import registry
from lavis.datasets.data_utils import load_video
from lavis.processors import transforms_video
from lavis.processors.base_processor import BaseProcessor
from lavis.processors.randaugment import VideoRandomAugment
from lavis.processors import functional_video as F
from omegaconf import OmegaConf
from torchvision import transforms
MAX_INT = registry.get("MAX_INT")
class AlproVideoBaseProcessor(BaseProcessor):
def __init__(self, mean=None, std=None, n_frms=MAX_INT):
if mean is None:
mean = (0.48145466, 0.4578275, 0.40821073)
if std is None:
std = (0.26862954, 0.26130258, 0.27577711)
self.normalize = transforms_video.NormalizeVideo(mean, std)
self.n_frms = n_frms
class ToUint8(object):
def __init__(self):
pass
def __call__(self, tensor):
return tensor.to(torch.uint8)
def __repr__(self):
return self.__class__.__name__
class ToTHWC(object):
"""
Args:
clip (torch.tensor, dtype=torch.uint8): Size is (C, T, H, W)
Return:
clip (torch.tensor, dtype=torch.float): Size is (T, H, W, C)
"""
def __init__(self):
pass
def __call__(self, tensor):
return tensor.permute(1, 2, 3, 0)
def __repr__(self):
return self.__class__.__name__
class ResizeVideo(object):
def __init__(self, target_size, interpolation_mode="bilinear"):
self.target_size = target_size
self.interpolation_mode = interpolation_mode
def __call__(self, clip):
"""
Args:
clip (torch.tensor): Video clip to be cropped. Size is (C, T, H, W)
Returns:
torch.tensor: central cropping of video clip. Size is
(C, T, crop_size, crop_size)
"""
return F.resize(clip, self.target_size, self.interpolation_mode)
def __repr__(self):
return self.__class__.__name__ + "(resize_size={0})".format(self.target_size)
@registry.register_processor("alpro_video_train")
class AlproVideoTrainProcessor(AlproVideoBaseProcessor):
def __init__(
self,
image_size=384,
mean=None,
std=None,
min_scale=0.5,
max_scale=1.0,
n_frms=MAX_INT,
):
super().__init__(mean=mean, std=std, n_frms=n_frms)
self.image_size = image_size
self.transform = transforms.Compose(
[
# Video size is (C, T, H, W)
transforms_video.RandomResizedCropVideo(
image_size,
scale=(min_scale, max_scale),
interpolation_mode="bicubic",
),
transforms_video.RandomHorizontalFlipVideo(),
ToTHWC(), # C, T, H, W -> T, H, W, C
VideoRandomAugment(
2,
5,
augs=[
"Identity",
"AutoContrast",
"Brightness",
"Sharpness",
"Equalize",
"ShearX",
"ShearY",
"TranslateX",
"TranslateY",
"Rotate",
],
),
ToUint8(),
transforms_video.ToTensorVideo(), # T, H, W, C -> C, T, H, W
self.normalize,
]
)
def __call__(self, vpath):
"""
Args:
clip (torch.tensor): Video clip to be cropped. Size is (C, T, H, W)
Returns:
torch.tensor: video clip after transforms. Size is (C, T, size, size).
"""
clip = load_video(
video_path=vpath,
n_frms=self.n_frms,
height=self.image_size,
width=self.image_size,
sampling="headtail",
)
return self.transform(clip)
@classmethod
def from_config(cls, cfg=None):
if cfg is None:
cfg = OmegaConf.create()
image_size = cfg.get("image_size", 256)
mean = cfg.get("mean", None)
std = cfg.get("std", None)
min_scale = cfg.get("min_scale", 0.5)
max_scale = cfg.get("max_scale", 1.0)
n_frms = cfg.get("n_frms", MAX_INT)
return cls(
image_size=image_size,
mean=mean,
std=std,
min_scale=min_scale,
max_scale=max_scale,
n_frms=n_frms,
)
@registry.register_processor("alpro_video_eval")
class AlproVideoEvalProcessor(AlproVideoBaseProcessor):
def __init__(self, image_size=256, mean=None, std=None, n_frms=MAX_INT):
super().__init__(mean=mean, std=std, n_frms=n_frms)
self.image_size = image_size
# Input video size is (C, T, H, W)
self.transform = transforms.Compose(
[
# frames will be resized during decord loading.
ToUint8(), # C, T, H, W
ToTHWC(), # T, H, W, C
transforms_video.ToTensorVideo(), # C, T, H, W
self.normalize, # C, T, H, W
]
)
def __call__(self, vpath):
"""
Args:
clip (torch.tensor): Video clip to be cropped. Size is (C, T, H, W)
Returns:
torch.tensor: video clip after transforms. Size is (C, T, size, size).
"""
clip = load_video(
video_path=vpath,
n_frms=self.n_frms,
height=self.image_size,
width=self.image_size,
)
return self.transform(clip)
@classmethod
def from_config(cls, cfg=None):
if cfg is None:
cfg = OmegaConf.create()
image_size = cfg.get("image_size", 256)
mean = cfg.get("mean", None)
std = cfg.get("std", None)
n_frms = cfg.get("n_frms", MAX_INT)
return cls(image_size=image_size, mean=mean, std=std, n_frms=n_frms)
|