File size: 8,695 Bytes
15e52d3
 
 
 
 
 
 
bb24dbf
15e52d3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
bb24dbf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
15e52d3
bb24dbf
15e52d3
 
 
bb24dbf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
15e52d3
bb24dbf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
15e52d3
 
 
 
 
 
 
 
a33fb40
 
 
15e52d3
 
 
 
 
a33fb40
15e52d3
 
 
 
 
 
 
 
a33fb40
15e52d3
 
 
 
bb24dbf
 
 
 
15e52d3
a33fb40
15e52d3
a33fb40
 
 
 
 
 
 
 
bb24dbf
15e52d3
bb24dbf
 
15e52d3
 
 
 
 
a33fb40
 
15e52d3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
bb24dbf
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
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
import gradio as gr
import clip
import torch
from qdrant_client import QdrantClient
import subprocess
import os
import uuid
import yt_dlp

# Setup CLIP
device = "cuda" if torch.cuda.is_available() else "cpu"
model, preprocess = clip.load("ViT-B/32", device=device)

# Setup Qdrant
client = QdrantClient(
    url="https://265484ec-5f64-40ec-a619-c7c9dffc2dd9.us-east-1-0.aws.cloud.qdrant.io:6333",
    api_key="eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJhY2Nlc3MiOiJtIn0.I2MgcVnOKkWmOXwFlqJqEqm6LFQIF4cjxU5up4wxwyw"
)
COLLECTION_NAME = "video_segments"

# Paths
CLIP_OUTPUT_DIR = "generated_clips"
os.makedirs(CLIP_OUTPUT_DIR, exist_ok=True)

# YouTube URLs mapping
VIDEO_URLS = {
    "temp_video_0.mp4": 'https://www.youtube.com/watch?v=9CGGh6ivg68',
    "temp_video_1.mp4": 'https://www.youtube.com/watch?v=WXoOohWU28Y', 
    "temp_video_2.mp4": 'https://www.youtube.com/watch?v=TV-DjM8242s', 
    "temp_video_3.mp4": 'https://www.youtube.com/watch?v=rCVlIVKqqGE',
    "temp_video_4.mp4": 'https://www.youtube.com/watch?v=lb_5AdUpfuA',
    "temp_video_5.mp4": 'https://www.youtube.com/watch?v=FCQ-rih6cHY', 
    "temp_video_6.mp4": 'https://www.youtube.com/watch?v=eQ6UE968Xe4', 
    "temp_video_7.mp4": 'https://www.youtube.com/watch?v=eFgkZKhNUdM'
}

DEFAULT_VIDEO_URL = VIDEO_URLS["temp_video_0.mp4"]

def extract_video_clip(video_url, start_time, end_time):
    """
    Use yt-dlp and ffmpeg to extract a clip directly from YouTube.
    """
    clip_name = f"clip_{uuid.uuid4().hex}.mp4"
    output_path = os.path.join(CLIP_OUTPUT_DIR, clip_name)
    duration = end_time - start_time
    
    print(f"[INFO] Attempting to extract clip from {video_url} ({start_time} - {end_time})")
    
    # Method 1: Use youtube-dl with ffmpeg
    try:
        # First, get the best direct URL from youtube-dl
        ydl_opts = {
            'format': 'bestvideo[ext=mp4]+bestaudio[ext=m4a]/best[ext=mp4]/best',
            'quiet': True
        }
        
        with yt_dlp.YoutubeDL(ydl_opts) as ydl:
            info = ydl.extract_info(video_url, download=False)
            formats = info.get('formats', [info])
            
            # Get the best format URL
            best_url = None
            for f in formats:
                if f.get('ext') == 'mp4' and f.get('url'):
                    best_url = f['url']
                    break
            
            if not best_url and info.get('url'):
                best_url = info['url']
                
            if not best_url:
                print("[WARN] Could not find a suitable direct URL")
                raise Exception("No suitable URL found")
                
            # Use ffmpeg to download the segment
            command = [
                "ffmpeg",
                "-ss", str(start_time),
                "-i", best_url,
                "-t", str(duration),
                "-c:v", "libx264",
                "-c:a", "aac",
                "-preset", "ultrafast",
                output_path,
                "-y"
            ]
            
            print(f"[INFO] Running ffmpeg command: {' '.join(command)}")
            result = subprocess.run(command, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
            
            if result.returncode != 0:
                print(f"[WARN] ffmpeg command failed with code {result.returncode}")
                print(f"[WARN] stderr: {result.stderr.decode('utf-8')}")
                raise Exception(f"ffmpeg failed with code {result.returncode}")
            
            if os.path.exists(output_path) and os.path.getsize(output_path) > 0:
                print(f"[INFO] Successfully extracted clip to {output_path}")
                return output_path
            else:
                print(f"[WARN] Output file missing or empty: {output_path}")
                raise Exception("Output file missing or empty")
                
    except Exception as e:
        print(f"[ERROR] Method 1 failed: {str(e)}")
    
    # Method 2: Download whole video first
    try:
        print("[INFO] Trying Method 2: Download full video first")
        temp_video = os.path.join(CLIP_OUTPUT_DIR, f"temp_{uuid.uuid4().hex}.mp4")
        
        ydl_opts = {
            'format': 'best[ext=mp4]/best',
            'outtmpl': temp_video,
            'quiet': True
        }
        
        with yt_dlp.YoutubeDL(ydl_opts) as ydl:
            ydl.download([video_url])
        
        if os.path.exists(temp_video) and os.path.getsize(temp_video) > 0:
            # Extract clip from the downloaded video
            command = [
                "ffmpeg",
                "-ss", str(start_time),
                "-i", temp_video,
                "-t", str(duration),
                "-c:v", "copy",
                "-c:a", "copy",
                output_path,
                "-y"
            ]
            
            print(f"[INFO] Running ffmpeg command: {' '.join(command)}")
            result = subprocess.run(command, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
            
            # Clean up temp file regardless of success
            try:
                os.remove(temp_video)
                print(f"[INFO] Removed temporary file {temp_video}")
            except Exception as cleanup_error:
                print(f"[WARN] Failed to remove temp file: {cleanup_error}")
            
            if result.returncode != 0:
                print(f"[WARN] ffmpeg command failed with code {result.returncode}")
                print(f"[WARN] stderr: {result.stderr.decode('utf-8')}")
                raise Exception(f"ffmpeg failed with code {result.returncode}")
            
            if os.path.exists(output_path) and os.path.getsize(output_path) > 0:
                print(f"[INFO] Successfully extracted clip to {output_path}")
                return output_path
            else:
                print(f"[WARN] Output file missing or empty: {output_path}")
                raise Exception("Output file missing or empty")
    
    except Exception as e:
        print(f"[ERROR] Method 2 failed: {str(e)}")
    
    # Both methods failed
    print("[ERROR] All extraction methods failed")
    return None

def time_to_seconds(time_str):
    h, m, s = time_str.split(':')
    return int(h) * 3600 + int(m) * 60 + float(s)

def search_and_clip_video(text_query: str):
    print(f"[INFO] Searching for: {text_query}")

    # Fixed-size wrapper
    wrapper = "<div style='width:100%; max-width:720px; height:405px; margin:auto;'>{}</div>"

    # Encode query
    with torch.no_grad():
        text_tokens = clip.tokenize([text_query]).to(device)
        text_features = model.encode_text(text_tokens)
        text_features /= text_features.norm(dim=1, keepdim=True)

    search_result = client.search(
        collection_name=COLLECTION_NAME,
        query_vector=text_features.cpu().numpy()[0].tolist(),
        limit=1,
    )

    if not search_result:
        print("[WARN] No result found.")
        return wrapper.format("<p style='text-align:center; padding-top:180px;'>No matching video found.</p>")

    hit = search_result[0]
    start = hit.payload.get("start", 0)
    end = hit.payload.get("end", 0)
    start = time_to_seconds(start) if isinstance(start, str) else float(start)
    end = time_to_seconds(end) if isinstance(end, str) else float(end)
    video_filename = hit.payload.get("video_path", "temp_video_0.mp4")
    video_url = VIDEO_URLS.get(video_filename, DEFAULT_VIDEO_URL)

    embed_url = video_url.replace("watch?v=", "embed/") + f"?start={int(start)}&end={int(end)}&autoplay=1"

    iframe = f"""
        <iframe width="100%" height="100%" 
                src="{embed_url}" 
                frameborder="0" 
                allow="autoplay; encrypted-media" 
                allowfullscreen></iframe>
    """
    return wrapper.format(iframe)
# Function to get a test video
def get_test_video():
    print("[INFO] Returning test YouTube URL")
    return DEFAULT_VIDEO_URL

# Gradio Interfaces
search_demo = gr.Interface(
    fn=search_and_clip_video,
    inputs=gr.Textbox(label="Enter search query", value="sample query"),
    # outputs=gr.Video(label="Video Result"),\
    outputs=gr.HTML(label="YouTube Clip"),
    title="🎥 Semantic Video Search with Clip Extraction",
    description="Returns a clipped video segment matching your query."
)

test_demo = gr.Interface(
    fn=get_test_video,
    inputs=None,
    outputs=gr.Video(label="Test Video"),
    title="Simple Video Test",
    description="Always displays the default video to verify video player works."
)

demo = gr.TabbedInterface(
    [search_demo, test_demo],
    ["Search Video", "Test Video Player"]
)

if __name__ == "__main__":
    demo.launch(share=True)