File size: 6,450 Bytes
19b3da3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from re import L

import cv2
import numpy as np
import torch
import torch.nn as nn
from transformers import CLIPConfig, CLIPVisionModel, PreTrainedModel

from internals.pipelines.commons import AbstractPipeline
from internals.util.config import get_nsfw_access, get_nsfw_threshold


def cosine_distance(image_embeds, text_embeds):
    normalized_image_embeds = nn.functional.normalize(image_embeds)
    normalized_text_embeds = nn.functional.normalize(text_embeds)
    return torch.mm(normalized_image_embeds, normalized_text_embeds.t())


class SafetyChecker:
    def load(self):
        self.model = StableDiffusionSafetyCheckerV2.from_pretrained(
            "CompVis/stable-diffusion-safety-checker", torch_dtype=torch.float16
        ).to("cuda")

    def apply(self, pipeline: AbstractPipeline):
        if hasattr(pipeline, "pipe"):
            pipeline.pipe.safety_checker = self.model
        if hasattr(pipeline, "pipe2"):
            pipeline.pipe2.safety_checker = self.model


class StableDiffusionSafetyCheckerV2(PreTrainedModel):
    config_class = CLIPConfig

    _no_split_modules = ["CLIPEncoderLayer"]

    def __init__(self, config: CLIPConfig):
        super().__init__(config)

        self.vision_model = CLIPVisionModel(config.vision_config)
        self.visual_projection = nn.Linear(
            config.vision_config.hidden_size, config.projection_dim, bias=False
        )

        self.concept_embeds = nn.Parameter(
            torch.ones(17, config.projection_dim), requires_grad=False
        )
        self.special_care_embeds = nn.Parameter(
            torch.ones(3, config.projection_dim), requires_grad=False
        )

        self.concept_embeds_weights = nn.Parameter(torch.ones(17), requires_grad=False)
        self.special_care_embeds_weights = nn.Parameter(
            torch.ones(3), requires_grad=False
        )

    @torch.no_grad()
    def forward(self, clip_input, images):
        pooled_output = self.vision_model(clip_input)[1]  # pooled_output
        image_embeds = self.visual_projection(pooled_output)

        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16
        special_cos_dist = (
            cosine_distance(image_embeds, self.special_care_embeds)
            .cpu()
            .float()
            .numpy()
        )
        cos_dist = (
            cosine_distance(image_embeds, self.concept_embeds).cpu().float().numpy()
        )

        result = []
        batch_size = image_embeds.shape[0]
        for i in range(batch_size):
            result_img = {
                "special_scores": {},
                "special_care": [],
                "concept_scores": {},
                "bad_concepts": [],
            }

            # increase this value to create a stronger `nfsw` filter
            # at the cost of increasing the possibility of filtering benign images
            adjustment = 0.0

            for concept_idx in range(len(special_cos_dist[0])):
                concept_cos = special_cos_dist[i][concept_idx]
                concept_threshold = self.special_care_embeds_weights[concept_idx].item()
                result_img["special_scores"][concept_idx] = round(
                    concept_cos - concept_threshold + adjustment, 3
                )
                if result_img["special_scores"][concept_idx] > 0:
                    result_img["special_care"].append(
                        {concept_idx, result_img["special_scores"][concept_idx]}
                    )
                    adjustment = 0.01

            for concept_idx in range(len(cos_dist[0])):
                concept_cos = cos_dist[i][concept_idx]
                concept_threshold = self.concept_embeds_weights[concept_idx].item()
                result_img["concept_scores"][concept_idx] = round(
                    concept_cos - concept_threshold + adjustment, 3
                )
                if result_img["concept_scores"][concept_idx] > get_nsfw_threshold():
                    result_img["bad_concepts"].append(concept_idx)

            result.append(result_img)

        has_nsfw_concepts = [len(res["bad_concepts"]) > 0 for res in result]

        # Blur images based on NSFW score
        # -------------------------------
        for idx, has_nsfw_concept in enumerate(has_nsfw_concepts):
            if any(has_nsfw_concepts) and not get_nsfw_access():
                if torch.is_tensor(images) or torch.is_tensor(images[0]):
                    image = images[idx].cpu().numpy().astype(np.float32)
                    image = cv2.blur(image, (30, 30))
                    image = torch.from_numpy(image)
                    images[idx] = image
                else:
                    images[idx] = cv2.blur(images[idx], (30, 30))

        if any(has_nsfw_concepts):
            print("NSFW")

        return images, has_nsfw_concepts

    @torch.no_grad()
    def forward_onnx(self, clip_input: torch.FloatTensor, images: torch.FloatTensor):
        pooled_output = self.vision_model(clip_input)[1]  # pooled_output
        image_embeds = self.visual_projection(pooled_output)

        special_cos_dist = cosine_distance(image_embeds, self.special_care_embeds)
        cos_dist = cosine_distance(image_embeds, self.concept_embeds)

        # increase this value to create a stronger `nsfw` filter
        # at the cost of increasing the possibility of filtering benign images
        adjustment = 0.0

        special_scores = (
            special_cos_dist - self.special_care_embeds_weights + adjustment
        )
        # special_scores = special_scores.round(decimals=3)
        special_care = torch.any(special_scores > 0, dim=1)
        special_adjustment = special_care * 0.01
        special_adjustment = special_adjustment.unsqueeze(1).expand(
            -1, cos_dist.shape[1]
        )

        concept_scores = (cos_dist - self.concept_embeds_weights) + special_adjustment
        # concept_scores = concept_scores.round(decimals=3)
        has_nsfw_concepts = torch.any(concept_scores > get_nsfw_threshold(), dim=1)

        # Blur images based on NSFW score
        # -------------------------------
        if not get_nsfw_access():
            image = images[has_nsfw_concepts].cpu().numpy().astype(np.float32)
            image = cv2.blur(image, (30, 30))
            image = torch.from_numpy(image)
            images[has_nsfw_concepts] = image

        return images, has_nsfw_concepts