File size: 4,727 Bytes
cd10cf2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9a9792d
 
cd10cf2
 
 
9a9792d
cd10cf2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a48bf14
07ae18d
a48bf14
 
 
 
 
 
 
 
b68d4a2
41332b4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ecf0e06
a48bf14
 
b68d4a2
a48bf14
9a9792d
cd10cf2
 
 
 
 
7233a34
e91f526
 
cd10cf2
 
 
 
 
e91f526
a48bf14
cd10cf2
 
 
 
 
 
 
 
 
 
 
e91f526
cd10cf2
 
e91f526
cd10cf2
 
 
 
7233a34
e91f526
7233a34
cd10cf2
 
 
 
 
 
7233a34
cd10cf2
 
 
 
 
 
 
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
import data
import torch
import gradio as gr
from models import imagebind_model
from models.imagebind_model import ModalityType


device = "cuda:0" if torch.cuda.is_available() else "cpu"
model = imagebind_model.imagebind_huge(pretrained=True)
model.eval()
model.to(device)


def image_text_zeroshot(image, text_list):
    image_paths = [image]
    labels = [label.strip(" ") for label in text_list.strip(" ").split("|")]
    inputs = {
        ModalityType.TEXT: data.load_and_transform_text(labels, device),
        ModalityType.VISION: data.load_and_transform_vision_data(image_paths, device),
    }

    with torch.no_grad():
        embeddings = model(inputs)

    scores = (
        torch.softmax(
            embeddings[ModalityType.VISION] @ embeddings[ModalityType.TEXT].T, dim=-1
        )
        .squeeze(0)
        .tolist()
    )

    score_dict = {label: score for label, score in zip(labels, scores)}

    return score_dict


def audio_text_zeroshot(audio, text_list):
    audio_paths = [audio]
    labels = [label.strip(" ") for label in text_list.strip(" ").split("|")]
    inputs = {
        ModalityType.TEXT: data.load_and_transform_text(labels, device),
        ModalityType.AUDIO: data.load_and_transform_audio_data(audio_paths, device),
    }

    with torch.no_grad():
        embeddings = model(inputs)

    scores = (
        torch.softmax(
            embeddings[ModalityType.AUDIO] @ embeddings[ModalityType.TEXT].T, dim=-1
        )
        .squeeze(0)
        .tolist()
    )

    score_dict = {label: score for label, score in zip(labels, scores)}

    return score_dict


def video_text_zeroshot(image, text_list):
    image_paths = [image]
    labels = [label.strip(" ") for label in text_list.strip(" ").split("|")]
    inputs = {
        ModalityType.TEXT: data.load_and_transform_text(labels, device),
        ModalityType.VISION: data.load_and_transform_vision_data(image_paths, device),
    }

    with torch.no_grad():
        embeddings = model(inputs)

    scores = (
        torch.softmax(
            embeddings[ModalityType.VISION] @ embeddings[ModalityType.TEXT].T, dim=-1
        )
        .squeeze(0)
        .tolist()
    )

    score_dict = {label: score for label, score in zip(labels, scores)}

    return score_dict

def doubleimage_text_zeroshot(image, image2,  text_list):
    image_paths = [image, image2]
    labels = [label.strip(" ") for label in text_list.strip(" ").split("|")]
    inputs = {
        ModalityType.TEXT: data.load_and_transform_text(labels, device),
        ModalityType.VISION: data.load_and_transform_vision_data(image_paths, device),
    }

    with torch.no_grad():
        embeddings = model(inputs)
 
    scores = (
        torch.softmax(
            embeddings[ModalityType.VISION] @ embeddings[ModalityType.TEXT].T, dim=-1
        )
        .squeeze(0)
        .tolist()
    )

    score_dict = {label: score for label, score in zip(labels, scores)}

    return score_dict



def doubleimage_text_zeroshotOLD(image, image2,  text_list):
    image_paths = [image, image2]
    labels = [label.strip(" ") for label in text_list.strip(" ").split("|")]
    inputs = {
        ModalityType.TEXT: data.load_and_transform_text(labels, device),
        ModalityType.VISION: data.load_and_transform_vision_data(image_paths, device),
    }

    with torch.no_grad():
        embeddings = model(inputs)
 



    return        str(torch.softmax(embeddings[ModalityType.VISION] @ embeddings[ModalityType.TEXT].T, dim=-1)   ) 

    

def inference(
    task,
    text_list=None,
    image=None,
    audio=None,
    image2=None,

):
    if task == "image-text":
        result = image_text_zeroshot(image, text_list)
    elif task == "audio-text":
        result = audio_text_zeroshot(audio, text_list)
    elif task == "embeddings":
        result = doubleimage_text_zeroshot(image, image2, text_list)
    else:
        raise NotImplementedError
    return result


def main():
    inputs = [
        gr.inputs.Radio(
            choices=[
                "image-text",
                "audio-text",
                "embeddings",
            ],
            type="value",
            default="embeddings",
            label="Task",
        ),
        gr.inputs.Textbox(lines=1, label="Candidate texts"),
        gr.inputs.Image(type="filepath", label="Input image"),
        gr.inputs.Audio(type="filepath", label="Input audio"),
        gr.inputs.Image(type="filepath", label="Input image2"),
    
    ]

    iface = gr.Interface(
        inference,
        inputs,
        "label",
        title="Multimodal AI assitive agents for Learning Disorders : Demo with embeddings of ImageBind: ",
    )

    iface.launch()


if __name__ == "__main__":
    main()