File size: 6,679 Bytes
1b92e8f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2821e52
1b92e8f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d833a17
1b92e8f
 
 
 
d833a17
1b92e8f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
56d047b
1b92e8f
 
5aeb32a
56d047b
 
5aeb32a
 
1b92e8f
 
 
 
56d047b
 
bc16d21
1b92e8f
 
 
 
 
 
 
 
56d047b
1b92e8f
 
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
import torch
from diffusers.loaders import AttnProcsLayers
from transformers import CLIPTextModel, CLIPTokenizer
from modules.beats.BEATs import BEATs, BEATsConfig
from modules.AudioToken.embedder import FGAEmbedder
from diffusers import AutoencoderKL, UNet2DConditionModel
from diffusers.models.attention_processor import LoRAAttnProcessor
from diffusers import StableDiffusionPipeline
import numpy as np
import gradio as gr


class AudioTokenWrapper(torch.nn.Module):
    """Simple wrapper module for Stable Diffusion that holds all the models together"""

    def __init__(
        self,
        lora,
        device,
    ):

        super().__init__()
        # Load scheduler and models
        self.tokenizer = CLIPTokenizer.from_pretrained(
            "CompVis/stable-diffusion-v1-4", subfolder="tokenizer"
        )
        self.text_encoder = CLIPTextModel.from_pretrained(
            "CompVis/stable-diffusion-v1-4", subfolder="text_encoder", revision=None
        )
        self.unet = UNet2DConditionModel.from_pretrained(
            "CompVis/stable-diffusion-v1-4", subfolder="unet", revision=None
        )
        self.vae = AutoencoderKL.from_pretrained(
            "CompVis/stable-diffusion-v1-4", subfolder="vae", revision=None
        )

        checkpoint = torch.load(
            'models/BEATs_iter3_plus_AS2M_finetuned_on_AS2M_cpt2.pt')
        cfg = BEATsConfig(checkpoint['cfg'])
        self.aud_encoder = BEATs(cfg)
        self.aud_encoder.load_state_dict(checkpoint['model'])
        self.aud_encoder.predictor = None
        input_size = 768 * 3
        self.embedder = FGAEmbedder(input_size=input_size, output_size=768)

        self.vae.eval()
        self.unet.eval()
        self.text_encoder.eval()
        self.aud_encoder.eval()

        if lora:
            # Set correct lora layers
            lora_attn_procs = {}
            for name in self.unet.attn_processors.keys():
                cross_attention_dim = None if name.endswith(
                    "attn1.processor") else self.unet.config.cross_attention_dim
                if name.startswith("mid_block"):
                    hidden_size = self.unet.config.block_out_channels[-1]
                elif name.startswith("up_blocks"):
                    block_id = int(name[len("up_blocks.")])
                    hidden_size = list(reversed(self.unet.config.block_out_channels))[block_id]
                elif name.startswith("down_blocks"):
                    block_id = int(name[len("down_blocks.")])
                    hidden_size = self.unet.config.block_out_channels[block_id]

                lora_attn_procs[name] = LoRAAttnProcessor(hidden_size=hidden_size,
                                                          cross_attention_dim=cross_attention_dim)

            self.unet.set_attn_processor(lora_attn_procs)
            self.lora_layers = AttnProcsLayers(self.unet.attn_processors)
            self.lora_layers.eval()
            lora_layers_learned_embeds = 'models/lora_layers_learned_embeds.bin'
            self.lora_layers.load_state_dict(torch.load(lora_layers_learned_embeds, map_location=device))
            self.unet.load_attn_procs(lora_layers_learned_embeds)

        self.embedder.eval()
        embedder_learned_embeds = 'models/embedder_learned_embeds.bin'
        self.embedder.load_state_dict(torch.load(embedder_learned_embeds, map_location=device))

        self.placeholder_token = '<*>'
        num_added_tokens = self.tokenizer.add_tokens(self.placeholder_token)
        if num_added_tokens == 0:
            raise ValueError(
                f"The tokenizer already contains the token {self.placeholder_token}. Please pass a different"
                " `placeholder_token` that is not already in the tokenizer."
            )
        self.placeholder_token_id = self.tokenizer.convert_tokens_to_ids(self.placeholder_token)
        # Resize the token embeddings as we are adding new special tokens to the tokenizer
        self.text_encoder.resize_token_embeddings(len(self.tokenizer))


def greet(audio):
    audio = audio[-1].astype(np.float32, order='C') / 32768.0
    weight_dtype = torch.float32
    prompt = 'a photo of <*>'

    audio_values = torch.unsqueeze(torch.tensor(audio), dim=0).to(device).to(dtype=weight_dtype)
    aud_features = model.aud_encoder.extract_features(audio_values)[1]
    audio_token = model.embedder(aud_features)

    token_embeds = model.text_encoder.get_input_embeddings().weight.data
    token_embeds[model.placeholder_token_id] = audio_token.clone()

    pipeline = StableDiffusionPipeline.from_pretrained(
        "CompVis/stable-diffusion-v1-4",
        tokenizer=model.tokenizer,
        text_encoder=model.text_encoder,
        vae=model.vae,
        unet=model.unet,
    ).to(device)
    image = pipeline(prompt, num_inference_steps=50, guidance_scale=7.5).images[0]
    return image


if __name__ == "__main__":

    lora = True
    device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
    model = AudioTokenWrapper(lora, device)

    description = """<p>
    This is a demo of <a href='https://pages.cs.huji.ac.il/adiyoss-lab/AudioToken' target='_blank'>AudioToken: Adaptation of Text-Conditioned Diffusion Models for Audio-to-Image Generation</a>.<br><br>
    In recent years, image generation has shown a great leap in performance, where diffusion models play a central role. Although generating high-quality images, such models are mainly conditioned on textual descriptions. This begs the question: "how can we adopt such models to be conditioned on other modalities?". We propose a novel method utilizing latent diffusion models trained for text-to-image-generation to generate images conditioned on audio recordings. Using a pre-trained audio encoding model, the proposed method encodes audio into a new token, which can be considered as an adaptation layer between the audio and text representations. Such a modeling paradigm requires a small number of trainable parameters, making the proposed approach appealing for lightweight optimization.<br><br>
    For more information, please see the original <a href='https://arxiv.org/abs/2305.13050' target='_blank'>paper</a> and <a href='https://github.com/guyyariv/AudioToken' target='_blank'>repo</a>.
    </p>"""

    examples = [
        ["assets/train.wav"],
        ["assets/dog barking.wav"],
        ["assets/airplane.wav"],
        ["assets/electric guitar.wav"],
        ["assets/female sings.wav"],
    ]

    demo = gr.Interface(
        fn=greet,
        inputs="audio",
        outputs="image",
        title='AudioToken',
        description=description,
        examples=examples
    )
    demo.launch()