File size: 5,694 Bytes
f1c3bf7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import os
import torch
import numpy as np
import torchvision
from urllib.request import urlopen
from PIL import Image, ImageDraw, ImageFont
from torchvision.transforms.functional import InterpolationMode
import torchvision.transforms as transforms
from decord import VideoReader

def get_font():
    truetype_url = 'https://cdn-lfs-us-1.huggingface.co/repos/19/7a/197a751ef710da1639736f1b5c9ebc26bd38d236aba7f10bcf8b553084c66907/336a838f4a78e150826be608dae69de59d50948c3d2b71760e096ae764154bdc?response-content-disposition=inline%3B+filename*%3DUTF-8%27%27SimHei.ttf%3B+filename%3D%22SimHei.ttf%22%3B&response-content-type=font%2Fttf&Expires=1720275312&Policy=eyJTdGF0ZW1lbnQiOlt7IkNvbmRpdGlvbiI6eyJEYXRlTGVzc1RoYW4iOnsiQVdTOkVwb2NoVGltZSI6MTcyMDI3NTMxMn19LCJSZXNvdXJjZSI6Imh0dHBzOi8vY2RuLWxmcy11cy0xLmh1Z2dpbmdmYWNlLmNvL3JlcG9zLzE5LzdhLzE5N2E3NTFlZjcxMGRhMTYzOTczNmYxYjVjOWViYzI2YmQzOGQyMzZhYmE3ZjEwYmNmOGI1NTMwODRjNjY5MDcvMzM2YTgzOGY0YTc4ZTE1MDgyNmJlNjA4ZGFlNjlkZTU5ZDUwOTQ4YzNkMmI3MTc2MGUwOTZhZTc2NDE1NGJkYz9yZXNwb25zZS1jb250ZW50LWRpc3Bvc2l0aW9uPSomcmVzcG9uc2UtY29udGVudC10eXBlPSoifV19&Signature=aZAXME5llGK90xUsPHRuWouco5T92ngs63hhW0gIAWmrUup4Ed5y4lSqB5khoLCLlMHK5lC4QJ58JTFFnmVFgFsKA-XfggYJLXu-TIC6DnvQCLz4L6EvLwCR05jzWOWn3trDorazP%7Enb8nuYKPgwGkpsukvCcqpx5Y0%7EfA4XsUCmcaddmkhFkkS1Wp2QWDnJjFGkuRnm8fQLW%7EG3JCdd7EyBkr2uWG%7E3W7ff62l-f%7EQTvtXIpYTHF3SAeqbB-DYQMUIbQJTuSs0TiQPt3WYvchrbuKN0aqR5OLvDJI2Fl0omJCL-wESyj9L%7EC2sCyY2LCDoE8b6-omgbQal2KHv7cA__&Key-Pair-Id=K24J24Z295AEI9'
    ff = urlopen(truetype_url)
    font = ImageFont.truetype(ff, size=40)
    return font

def padding_336(b, pad=336):
    width, height = b.size
    tar = int(np.ceil(height / pad) * pad)
    top_padding = 0 # int((tar - height)/2)
    bottom_padding = tar - height - top_padding
    left_padding = 0
    right_padding = 0
    b = transforms.functional.pad(b, [left_padding, top_padding, right_padding, bottom_padding], fill=[255,255,255])

    return b

def Image_transform(img, hd_num=25):
    width, height = img.size
    trans = False
    if width < height:
        img = img.transpose(Image.TRANSPOSE)
        trans = True
        width, height = img.size
    ratio = (width/ height)
    scale = 1
    while scale*np.ceil(scale/ratio) <= hd_num:
        scale += 1
    scale -= 1
    scale = min(np.ceil(width / 560), scale)
    new_w = int(scale * 560)
    new_h = int(new_w / ratio)
    #print (scale, f'{height}/{new_h}, {width}/{new_w}')

    img = transforms.functional.resize(img, [new_h, new_w],)
    img = padding_336(img, 560)
    width, height = img.size
    if trans:
        img = img.transpose(Image.TRANSPOSE)

    return img


def Video_transform(img, hd_num=25):
    width, height = img.size
    trans = False
    if width < height:
        img = img.transpose(Image.TRANSPOSE)
        trans = True
        width, height = img.size
    ratio = (width/ height)
    scale = 1
    new_h = int(scale * 560)
    new_w = int(new_h * ratio)
    #print (new_h, new_w)

    img = transforms.functional.resize(img, [new_h, new_w],)
    img = img.transpose(Image.TRANSPOSE)
    img = padding_336(img, 560)
    width, height = img.size
    if not trans:
        img = img.transpose(Image.TRANSPOSE)

    return img

def frame2img(imgs, font):
    new_imgs = []
    for img in imgs:
        w, h = img.size
        scale = w/h
        if w > h:
            new_w = 560 * 2
            new_h = int(560 * 2 / scale)
        else:
            new_w = int(560 * 2 * scale)
            new_h = 560 * 2
        img = transforms.functional.resize(img, [new_h, new_w],)
        new_imgs.append(img)
    imgs = new_imgs
    new_w = 0
    new_h = 0
    pad = 40
    if w > h:
        for im in imgs:
            w,h = im.size
            new_w = max(new_w, w)
            new_h += h + 10 + pad
        new_img = Image.new('RGB', (new_w, new_h), 'white')
        draw = ImageDraw.Draw(new_img)
        curr_h = 0
        for idx, im in enumerate(imgs):
            w,h = im.size
            new_img.paste(im, (0, pad + curr_h))
            draw.text((0, curr_h ), f'<IMAGE {idx}>', font=font, fill='black')
            if idx + 1 < len(imgs):
                draw.line([(0, pad +curr_h + h +5), (new_w, pad +curr_h + h +5)], fill = 'black', width=2)
            curr_h += h + 10 + pad
        #print (new_w, new_h)
    else:
        for im in imgs:
            w,h = im.size
            new_w += w + 10
            new_h = max(new_h, h)
        new_h += pad
        new_img = Image.new('RGB', (new_w, new_h), 'white')
        draw = ImageDraw.Draw(new_img)
        curr_w = 0
        for idx, im in enumerate(imgs):
            w,h = im.size
            new_img.paste(im, (curr_w, pad))
            draw.text((curr_w, 0), f'<IMAGE {idx}>', font=font, fill='black')
            if idx + 1 < len(imgs):
                draw.line([(curr_w + w + 5, 0), (curr_w + w + 5, new_h)], fill = 'black', width=2)
            curr_w += w + 10
    return new_img

def load_video(video_path, num_frm=32, start=None, end=None):
    vid = VideoReader(video_path, num_threads=1)
    fps = vid.get_avg_fps()
    t_stride = int(round(float(fps) / int(1)))
    start_idx = 0 if start is None else start
    end_idx = len(vid) if end is None else end
    all_pos = list(range(start_idx, end_idx, t_stride))
    try:
        images = [vid[i].numpy() for i in all_pos]
    except:
        images = [vid[i].asnumpy() for i in all_pos]
    if len(images) > num_frm:
        num_frm = min(num_frm, len(images))
        step_size = len(images) / (num_frm + 1)
        indices = [int(i*step_size) for i in range(num_frm)]
        images = [images[i] for i in indices]
    images = [Image.fromarray(arr) for arr in images]
    return images