File size: 7,209 Bytes
025f5af
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
fd87968
025f5af
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch
import torch.nn as nn

from transformers import CLIPVisionModel, CLIPImageProcessor, CLIPVisionConfig


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='cuda')# Error on the HuggingFace for auto
        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
    
    def feature_select_withcls(self, image_forward_outs):
        image_features = image_forward_outs.hidden_states[self.select_layer]
        image_features = image_features
        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_withcls(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_withcls(image_forward_outs).to(images.dtype)

        return image_features
    def forward_select(self, images, token_num):
        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_withcls(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, output_attentions=True)
            attn_weights  = image_forward_outs.attentions[-2]
            hidden_states = image_forward_outs.hidden_states[-2]
            dominant_num =  token_num

            ## Dominant Visual Tokens
            cls_idx = 0
            cls_attention = attn_weights[:, :, cls_idx, cls_idx+1:]  
            cls_attention_sum = cls_attention.sum(dim=1)  
            topk_indices = cls_attention_sum.topk(dominant_num, dim=1).indices 
            
            topk_indices_sorted = torch.sort(topk_indices, dim=1).values

        return topk_indices_sorted
    @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



class CLIPVisionTowerS2(CLIPVisionTower):
    def __init__(self, vision_tower, args, delay_load=False):
        super().__init__(vision_tower, args, delay_load)

        self.s2_scales = getattr(args, 's2_scales', '336,672,1008')
        self.s2_scales = list(map(int, self.s2_scales.split(',')))
        self.s2_scales.sort()
        self.s2_split_size = self.s2_scales[0]
        self.s2_image_size = self.s2_scales[-1]

        try:
            from s2wrapper import forward as multiscale_forward
        except ImportError:
            raise ImportError('Package s2wrapper not found! Please install by running: \npip install git+https://github.com/bfshi/scaling_on_scales.git')
        self.multiscale_forward = multiscale_forward

        # change resize/crop size in preprocessing to the largest image size in s2_scale
        if not delay_load or getattr(args, 'unfreeze_mm_vision_tower', False):
            self.image_processor.size['shortest_edge'] = self.s2_image_size
            self.image_processor.crop_size['height'] = self.image_processor.crop_size['width'] = self.s2_image_size

    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.image_processor.size['shortest_edge'] = self.s2_image_size
        self.image_processor.crop_size['height'] = self.image_processor.crop_size['width'] = self.s2_image_size

        self.is_loaded = True

    @torch.no_grad()
    def forward_feature(self, images):
        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

    @torch.no_grad()
    def forward(self, images):
        if type(images) is list:
            image_features = []
            for image in images:
                image_feature = self.multiscale_forward(self.forward_feature, image.unsqueeze(0), img_sizes=self.s2_scales, max_split_size=self.s2_split_size)
                image_features.append(image_feature)
        else:
            image_features = self.multiscale_forward(self.forward_feature, images, img_sizes=self.s2_scales, max_split_size=self.s2_split_size)

        return image_features

    @property
    def hidden_size(self):
        return self.config.hidden_size * len(self.s2_scales)