File size: 6,281 Bytes
f799d96
 
 
b0a48de
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f799d96
 
 
 
 
 
 
 
 
 
 
 
 
987b643
 
 
 
 
f799d96
 
 
 
987b643
f799d96
 
 
b0a48de
f799d96
987b643
f799d96
b0a48de
f799d96
 
 
 
 
 
b0a48de
f799d96
 
 
 
 
 
 
 
 
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
import gradio as gr
import cv2
import os
import spaces
import tempfile
from torchvision import transforms
from torchvision.transforms import Compose
import torch
import numpy as np
from PIL import Image
import torch.nn.functional as F
from pytorchvideo.transforms.functional import predict_depth
from transformers import pipeline, TimesformerModel, VideoMAEImageProcessor
from utils import *
from algorithm import *

@spaces.GPU
def make_video(video_path, outdir='./summarized_video',encoder='Kmeans'):
    if encoder not in ["Kmeans", "Sum of Squared Difference 01", "Sum of Squared Difference 02"]:
        encoder = "Kmeans"
    # nen them vao cac truong hop mo hinh khac
    margin_width = 50

    model, processor, device = load_model()

    # total_params = sum(param.numel() for param in model.parameters())
    # print('Total parameters: {:.2f}M'.format(total_params / 1e6))

    if os.path.isfile(video_path):
        if video_path.endswith('txt'):
            with open(video_path, 'r') as f:
                lines = f.read().splitlines()
        else:
            filenames = [video_path]
    else:
        filenames = os.listdir(video_path)
        filenames = [os.path.join(video_path, filename) for filename in filenames if not filename.startswith('.')]
        filenames.sort()
    
    for k, filename in enumerate(filenames):
        print('Progress {:}/{:},'.format(k+1, len(filenames)), 'Processing', filename)
        
        raw_video = cv2.VideoCapture(filename)
        frame_width, frame_height = int(raw_video.get(cv2.CAP_PROP_FRAME_WIDTH)), int(raw_video.get(cv2.CAP_PROP_FRAME_HEIGHT))
        frame_rate = int(raw_video.get(cv2.CAP_PROP_FPS))
        #length = int(raw_video.get(cv2.CAP_PROP_FRAME_COUNT))
        output_width = frame_width * 2 + margin_width
        
        filename = os.path.basename(filename)

        # Find the size to resize
        if "shortest_edge" in processor.size:
            height = width = processor.size["shortest_edge"]
        else:
            height = processor.size["height"]
            width = processor.size["width"] 
        resize_to = (height, width)

        # F/Fs
        clip_sample_rate = 1
        # F
        num_frames = 8

        frames = []
        features = []

        # output_path = os.path.join(outdir, filename[:filename.rfind('.')] + '_video_depth.mp4')
        with tempfile.NamedTemporaryFile(delete=False, suffix='.mp4') as tmpfile:
            output_path = tmpfile.name
        #out = cv2.VideoWriter(output_path, cv2.VideoWriter_fourcc(*"avc1"), frame_rate, (output_width, frame_height))
        fourcc = cv2.VideoWriter_fourcc(*'mp4v')
        out = cv2.VideoWriter(output_path, fourcc, frame_rate, (output_width, frame_height))
        # count=0
        
        while raw_video.isOpened():
            ret, raw_frame = raw_video.read()
            if not ret:
                break
            
            raw_frame = cv2.resize(raw_frame, resize_to)
            frames.append(raw_frame)

        # Find key frames by selecting frames with clip_sample_rate
        key_frames = frames[::clip_sample_rate] 
        #print('total of frames after sample:', len(selected_frames))

        # Remove redundant frames to make the number of frames can be divided by num_frames
        num_redudant_frames = len(key_frames) - (len(key_frames) % num_frames)

        # Final key frames
        final_key_frames = key_frames[:num_redudant_frames]
        #print('total of frames after remove redundant frames:', len(selected_frames))

        for i in range(0, len(final_key_frames), num_frames):
            if i % num_frames*50 == 0:
                print(f"Loading {i}/{len(final_key_frames)}")
        
        # Input clip to the model
        input_frames = final_key_frames[i:i+num_frames]
        # Extract features
        batch_features = extract_features(input_frames, device, model, processor)
        # Convert to numpy array to decrease the memory usage
        batch_features = np.array(batch_features.cpu().detach().numpy())
        features.extend(batch_features)

        number_of_clusters = round(len(features)*0.15)

        selected_frames = []
        if encoder == "Kmeans":
            selected_frames = kmeans(features, number_of_clusters)
        elif encoder == "Sum of Squared Difference 01":
            selected_frames = tt01(features, 400)
        else:
            selected_frames = tt02(features, 400)

        video_writer = cv2.VideoWriter(output_path, cv2.VideoWriter_fourcc(*'mp4v'), frame_rate, (frames[0].shape[1], frames[0].shape[0]))
        for idx in selected_frames:
            video_writer.write(frames[idx])
        
        raw_video.release()
        video_writer.release()
        print("Completed summarizing the video (wait for a moment to load).")
        return output_path

css = """
#img-display-container {
    max-height: 100vh;
    }
#img-display-input {
    max-height: 80vh;
    }
#img-display-output {
    max-height: 80vh;
    }
"""

title = "# Video Summarization Demo"
description = """Video Summarization using Timesformer.

Author: Nguyen Hoai Nam.
"""

with gr.Blocks(css=css) as demo:
    gr.Markdown(title)
    gr.Markdown(description)
    gr.Markdown("### Video Summarization demo")

    with gr.Row():
        input_video = gr.Video(label="Input Video")
        algorithm_type = gr.Dropdown(["Kmeans", "Sum of Squared Difference 01", "Sum of Squared Difference 02"], type="value", label='Algorithm')
    submit = gr.Button("Submit")
    processed_video = gr.Video(label="Summarized Video")

    def on_submit(uploaded_video,algorithm_type):
                
        # Process the video and get the path of the output video
        #output_video_path = make_video(uploaded_video,encoder=model_type)
        pass
        #return output_video_path

    submit.click(on_submit, inputs=[input_video, algorithm_type], outputs=processed_video)

    #example_files = os.listdir('assets/examples_video')
    #example_files.sort()
    #example_files = [os.path.join('assets/examples_video', filename) for filename in example_files]
    #examples = gr.Examples(examples=example_files, inputs=[input_video], outputs=processed_video, fn=on_submit, cache_examples=True)
    

if __name__ == '__main__':
    demo.queue().launch()