File size: 5,587 Bytes
b7c7aa0
 
dc465b0
b7c7aa0
 
dc465b0
 
 
 
 
 
b7c7aa0
 
 
 
 
dc465b0
b7c7aa0
 
 
 
 
 
 
 
 
 
 
 
dc465b0
b7c7aa0
 
dc465b0
 
 
 
 
 
50e7d4a
 
 
 
 
 
 
 
 
 
b7c7aa0
 
 
 
 
 
 
 
50e7d4a
 
1416e63
 
50e7d4a
 
 
 
 
 
 
 
b7c7aa0
 
 
50e7d4a
 
1416e63
 
50e7d4a
1416e63
 
50e7d4a
 
1416e63
b7c7aa0
 
dc465b0
 
b7c7aa0
 
 
 
 
50e7d4a
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
# app.py
import gradio as gr
from utils import VideoProcessor, AzureAPI, GoogleAPI, AnthropicAPI, OpenAIAPI
from constraint import SYS_PROMPT, USER_PROMPT

def fast_caption(sys_prompt, usr_prompt, temp, top_p, max_tokens, model, key, endpoint, video, frame_format, frame_limit):
    processor = VideoProcessor(frame_format=frame_format, frame_limit=frame_limit)
    frames = processor._decode(video)
    
    base64_list = processor.to_base64_list(frames)
    debug_image = processor.concatenate(frames)

    if not key or not endpoint:
        return "", f"API key or endpoint is missing. Processed {len(frames)} frames.", debug_image
    
    api = AzureAPI(key=key, endpoint=endpoint, model=model, temp=temp, top_p=top_p, max_tokens=max_tokens)
    caption = api.get_caption(sys_prompt, usr_prompt, base64_list)
    return f"{caption}", f"Using model '{model}' with {len(frames)} frames extracted.", debug_image

with gr.Blocks() as Core:
    with gr.Row(variant="panel"):
        with gr.Column(scale=6):
            with gr.Accordion("Debug", open=False):
                info = gr.Textbox(label="Info", interactive=False)
                frame = gr.Image(label="Frame", interactive=False)
            with gr.Accordion("Configuration", open=False):
                with gr.Row():
                    temp = gr.Slider(0, 1, 0.3, step=0.1, label="Temperature")
                    top_p = gr.Slider(0, 1, 0.75, step=0.1, label="Top-P")
                    max_tokens = gr.Slider(512, 4096, 1024, step=1, label="Max Tokens")
                with gr.Row():
                    frame_format = gr.Dropdown(label="Frame Format", value="JPEG", choices=["JPEG", "PNG"], interactive=False)
                    frame_limit = gr.Slider(1, 100, 10, step=1, label="Frame Limits")
            with gr.Tabs():
                with gr.Tab("User"):
                    usr_prompt = gr.Textbox(USER_PROMPT, label="User Prompt", lines=10, max_lines=100, show_copy_button=True)
                with gr.Tab("System"):
                    sys_prompt = gr.Textbox(SYS_PROMPT, label="System Prompt", lines=10, max_lines=100, show_copy_button=True)
            with gr.Tabs():
                with gr.Tab("Azure"):
                    result = gr.Textbox(label="Result", lines=15, max_lines=100, show_copy_button=True, interactive=False)
                with gr.Tab("Google"):
                    result_gg = gr.Textbox(label="Result", lines=15, max_lines=100, show_copy_button=True, interactive=False)
                with gr.Tab("Anthropic"):
                    result_ac = gr.Textbox(label="Result", lines=15, max_lines=100, show_copy_button=True, interactive=False)
                with gr.Tab("OpenAI"):
                    result_oai = gr.Textbox(label="Result", lines=15, max_lines=100, show_copy_button=True, interactive=False)

        with gr.Column(scale=2):
            with gr.Column():
                with gr.Accordion("Model Provider", open=True):
                    with gr.Tabs():
                        with gr.Tab("Azure"):
                            model = gr.Dropdown(label="Model", value="GPT-4o", choices=["GPT-4o", "GPT-4v"], interactive=False)
                            key = gr.Textbox(label="Azure API Key")
                            endpoint = gr.Textbox(label="Azure Endpoint")
                        with gr.Tab("Google"):
                            model_gg = gr.Dropdown(label="Model", value="Gemini-1.5-Flash", choices=["Gemini-1.5-Flash", "Gemini-1.5-Pro"], interactive=False)
                            key_gg = gr.Textbox(label="Gemini API Key")
                            endpoint_gg = gr.Textbox(label="Gemini API Endpoint")
                        with gr.Tab("Anthropic"):
                            model_ac = gr.Dropdown(label="Model", value="Claude-3-Opus", choices=["Claude-3-Opus", "Claude-3-Sonnet"], interactive=False)
                            key_ac = gr.Textbox(label="Anthropic API Key")
                            endpoint_ac = gr.Textbox(label="Anthropic Endpoint")
                        with gr.Tab("OpenAI"):
                            model_oai = gr.Dropdown(label="Model", value="GPT-4o", choices=["GPT-4o", "GPT-4v"], interactive=False)
                            key_oai = gr.Textbox(label="OpenAI API Key")
                            endpoint_oai = gr.Textbox(label="OpenAI Endpoint")
                with gr.Accordion("Data Source", open=True):
                    with gr.Tabs():
                        with gr.Tab("Upload"):
                            video_src = gr.Video(sources="upload", show_label=False, show_share_button=False, mirror_webcam=False)
                        with gr.Tab("HF"):
                            video_hf = gr.Text(label="Huggingface File Path")
                            video_hf_auth = gr.Text(label="Huggingface Token")
                        with gr.Tab("Onedrive"):
                            video_od = gr.Text("Microsoft Onedrive")
                            video_od_auth = gr.Text(label="Microsoft Onedrive Token")
                        with gr.Tab("Google Drive"):
                            video_gd = gr.Text()
                            video_gd_auth = gr.Text(label="Google Drive Access Token")
                caption_button = gr.Button("Caption", variant="primary", size="lg")
        caption_button.click(
            fast_caption, 
            inputs=[sys_prompt, usr_prompt, temp, top_p, max_tokens, model, key, endpoint, video_src, frame_format, frame_limit], 
            outputs=[result, info, frame]
        )

if __name__ == "__main__":
    Core.launch()