Spaces:
Sleeping
Sleeping
File size: 6,624 Bytes
c315863 |
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 |
# This file is modified from https://github.com/haotian-liu/LLaVA/
import torch
import torch.nn as nn
from transformers import (
CLIPVisionModel, CLIPImageProcessor, CLIPVisionConfig,
SiglipVisionModel, SiglipImageProcessor, SiglipVisionConfig
)
class CLIPVisionTower(nn.Module):
def __init__(self, vision_tower, args, delay_load=False):
super().__init__()
self.is_loaded = False
self.vision_tower_name = vision_tower
self.select_layer = args.mm_vision_select_layer
self.select_feature = getattr(args, 'mm_vision_select_feature', 'patch')
if not delay_load:
self.load_model()
elif getattr(args, 'unfreeze_mm_vision_tower', False):
self.load_model()
else:
self.cfg_only = CLIPVisionConfig.from_pretrained(self.vision_tower_name)
def load_model(self, device_map=None):
if self.is_loaded:
print('{} is already loaded, `load_model` called again, skipping.'.format(self.vision_tower_name))
return
self.image_processor = CLIPImageProcessor.from_pretrained(self.vision_tower_name)
self.vision_tower = CLIPVisionModel.from_pretrained(self.vision_tower_name, device_map=device_map)
self.vision_tower.requires_grad_(False)
self.is_loaded = True
def feature_select(self, image_forward_outs):
image_features = image_forward_outs.hidden_states[self.select_layer]
if self.select_feature == 'patch':
image_features = image_features[:, 1:]
elif self.select_feature == 'cls_patch':
image_features = image_features
else:
raise ValueError(f'Unexpected select feature: {self.select_feature}')
return image_features
@torch.no_grad()
def forward(self, images):
if type(images) is list:
image_features = []
for image in images:
image_forward_out = self.vision_tower(image.to(device=self.device, dtype=self.dtype).unsqueeze(0), output_hidden_states=True)
image_feature = self.feature_select(image_forward_out).to(image.dtype)
image_features.append(image_feature)
else:
image_forward_outs = self.vision_tower(images.to(device=self.device, dtype=self.dtype), output_hidden_states=True)
image_features = self.feature_select(image_forward_outs).to(images.dtype)
return image_features
@property
def dummy_feature(self):
return torch.zeros(1, self.hidden_size, device=self.device, dtype=self.dtype)
@property
def dtype(self):
return self.vision_tower.dtype
@property
def device(self):
return self.vision_tower.device
@property
def config(self):
if self.is_loaded:
return self.vision_tower.config
else:
return self.cfg_only
@property
def hidden_size(self):
return self.config.hidden_size
@property
def num_patches_per_side(self):
return self.config.image_size // self.config.patch_size
@property
def num_patches(self):
return (self.config.image_size // self.config.patch_size) ** 2
@property
def image_size(self):
return self.config.image_size
class SiglipVisionTower(nn.Module):
def __init__(self, vision_tower, args, delay_load=False):
super().__init__()
self.is_loaded = False
self.vision_tower_name = vision_tower
self.select_layer = args.mm_vision_select_layer
self.input_image_size = args.input_image_size
self.select_feature = getattr(args, 'mm_vision_select_feature', 'patch')
self.is_loaded = False
if not delay_load:
self.load_model()
else:
self.cfg_only = SiglipVisionConfig.from_pretrained(self.vision_tower_name)
def load_model(self, device_map=None):
if self. is_loaded:
return
self.image_processor = SiglipImageProcessor.from_pretrained(self.vision_tower_name)
self.vision_tower = SiglipVisionModel.from_pretrained(self.vision_tower_name)
self.image_processor.crop_size = {'height':self.input_image_size, 'width':self.input_image_size}
self.is_loaded = True
def feature_select(self, image_forward_outs, dtype):
image_features = image_forward_outs.hidden_states
if self.select_feature == 'patch':
image_features = image_features[self.select_layer].to(dtype)
elif self.select_feature == 'list':
image_features = [feature.to(dtype) for feature in image_features[::7]]
else:
raise ValueError(f'Unexpected select feature: {self.select_feature}')
return image_features
def forward(self, images):
if type(images) is list:
image_features = []
for image in images:
image_forward_out = self.vision_tower(image.to(device=self.device, dtype=self.dtype).unsqueeze(0), output_hidden_states=True)
image_feature = self.feature_select(image_forward_out, image.dtype)
image_features.append(image_feature)
else:
batch_size = images.shape[0]
chunk_size = 256
image_features = []
for i in range(0, batch_size, chunk_size):
chunk = images[i:i+chunk_size].to(device=self.device, dtype=self.dtype)
chunk_forward_outs = self.vision_tower(chunk, output_hidden_states=True)
chunk_features = self.feature_select(chunk_forward_outs, images.dtype)
image_features.append(chunk_features)
image_features = torch.cat(image_features, dim=0)
return image_features
@property
def dummy_feature(self):
return torch.zeros(1, self.hidden_size, device=self.device, dtype=self.dtype)
@property
def dtype(self):
return self.vision_tower.dtype
@property
def device(self):
return self.vision_tower.device
@property
def config(self):
if self.is_loaded:
return self.vision_tower.config
else:
return self.cfg_only
@property
def hidden_size(self):
return self.config.hidden_size
@property
def num_patches(self):
return (self.config.image_size // self.config.patch_size) ** 2
@property
def num_patches_per_side(self):
return self.config.image_size // self.config.patch_size
@property
def image_size(self):
return self.config.image_size |