File size: 6,749 Bytes
ee21b96
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0962f0f
 
ee21b96
 
 
 
 
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
import os

os.system('cd fairseq;'
          'pip install --use-feature=in-tree-build ./; cd ..')
os.system('ls -l')

import torch
import numpy as np
from fairseq import utils, tasks
from fairseq import checkpoint_utils
from utils.eval_utils import eval_step
from data.mm_data.ocr_dataset import ocr_resize
from tasks.mm_tasks.ocr import OcrTask
from PIL import Image, ImageDraw
from torchvision import transforms
from typing import List, Tuple
import cv2
from easyocrlite import ReaderLite
import gradio as gr


# Register refcoco task
tasks.register_task('ocr', OcrTask)

os.system('wget http://xc-models.oss-cn-zhangjiakou.aliyuncs.com/ofa/chinese/ocr/general/checkpoint_last.pt; '
          'mkdir -p checkpoints; mv checkpoint_last.pt checkpoints/ocr.pt')

# turn on cuda if GPU is available
use_cuda = torch.cuda.is_available()
# use fp16 only when GPU is available
use_fp16 = False

mean = [0.5, 0.5, 0.5]
std = [0.5, 0.5, 0.5]

Rect = Tuple[int, int, int, int]
FourPoint = Tuple[Tuple[int, int], Tuple[int, int], Tuple[int, int], Tuple[int, int]]


def four_point_transform(image: np.ndarray, rect: FourPoint) -> np.ndarray:
    (tl, tr, br, bl) = rect

    widthA = np.sqrt(((br[0] - bl[0]) ** 2) + ((br[1] - bl[1]) ** 2))
    widthB = np.sqrt(((tr[0] - tl[0]) ** 2) + ((tr[1] - tl[1]) ** 2))
    maxWidth = max(int(widthA), int(widthB))

    # compute the height of the new image, which will be the
    # maximum distance between the top-right and bottom-right
    # y-coordinates or the top-left and bottom-left y-coordinates
    heightA = np.sqrt(((tr[0] - br[0]) ** 2) + ((tr[1] - br[1]) ** 2))
    heightB = np.sqrt(((tl[0] - bl[0]) ** 2) + ((tl[1] - bl[1]) ** 2))
    maxHeight = max(int(heightA), int(heightB))

    dst = np.array(
        [[0, 0], [maxWidth - 1, 0], [maxWidth - 1, maxHeight - 1], [0, maxHeight - 1]],
        dtype="float32",
    )

    # compute the perspective transform matrix and then apply it
    M = cv2.getPerspectiveTransform(rect, dst)
    warped = cv2.warpPerspective(image, M, (maxWidth, maxHeight))

    return warped


def get_images(image_path: str, reader: ReaderLite, **kwargs):
    results = reader.process(image_path, **kwargs)
    return results


def draw_boxes(image, bounds, color='red', width=2):
    draw = ImageDraw.Draw(image)
    for bound in bounds:
        p0, p1, p2, p3 = bound
        draw.line([*p0, *p1, *p2, *p3, *p0], fill=color, width=width)
    return image


def encode_text(task, text, length=None, append_bos=False, append_eos=False):
    bos_item = torch.LongTensor([task.src_dict.bos()])
    eos_item = torch.LongTensor([task.src_dict.eos()])
    pad_idx = task.src_dict.pad()

    s = task.tgt_dict.encode_line(
        line=task.bpe.encode(text),
        add_if_not_exist=False,
        append_eos=False
    ).long()
    if length is not None:
        s = s[:length]
    if append_bos:
        s = torch.cat([bos_item, s])
    if append_eos:
        s = torch.cat([s, eos_item])
    return s


def patch_resize_transform(patch_image_size=480, is_document=False):
    _patch_resize_transform = transforms.Compose(
        [
            lambda image: ocr_resize(
                image, patch_image_size, is_document=is_document
            ),
            transforms.ToTensor(),
            transforms.Normalize(mean=mean, std=std),
        ]
    )
    
    return _patch_resize_transform


# Construct input for caption task
def construct_sample(task, image: Image, patch_image_size=480):
    bos_item = torch.LongTensor([task.src_dict.bos()])
    eos_item = torch.LongTensor([task.src_dict.eos()])
    pad_idx = task.src_dict.pad()

    patch_image = patch_resize_transform(patch_image_size)(image).unsqueeze(0)
    patch_mask = torch.tensor([True])
    src_text = encode_text(task, "图片上的文字是什么?", append_bos=True, append_eos=True).unsqueeze(0)
    src_length = torch.LongTensor([s.ne(pad_idx).long().sum() for s in src_text])
    sample = {
        "id":np.array(['42']),
        "net_input": {
            "src_tokens": src_text,
            "src_lengths": src_length,
            "patch_images": patch_image,
            "patch_masks": patch_mask,
        },
        "target": None
    }
    return sample


# Function to turn FP32 to FP16
def apply_half(t):
    if t.dtype is torch.float32:
        return t.to(dtype=torch.half)
    return t


def ocr(ckpt, img, out_img):
    reader = ReaderLite()
    overrides={"eval_cider":False, "beam":8, "max_len_b":128, "patch_image_size":480, "orig_patch_image_size":224, "no_repeat_ngram_size":0, "seed":7}
    models, cfg, task = checkpoint_utils.load_model_ensemble_and_task(
        utils.split_paths(ckpt),
        arg_overrides=overrides
    )

    # Move models to GPU
    for model in models:
        model.eval()
        if use_fp16:
            model.half()
        if use_cuda and not cfg.distributed_training.pipeline_model_parallel:
            model.cuda()
        model.prepare_for_inference_(cfg)

    # Initialize generator
    generator = task.build_generator(models, cfg.generation)

    bos_item = torch.LongTensor([task.src_dict.bos()])
    eos_item = torch.LongTensor([task.src_dict.eos()])
    pad_idx = task.src_dict.pad()

    orig_image = Image.open(img)
    results = get_images(img, reader)
    box_list, image_list = zip(*results)
    draw_boxes(orig_image, box_list)
    orig_image.save(out_img)

    ocr_result = []
    for box, image in zip(box_list, image_list):
        image = Image.fromarray(image)
        sample = construct_sample(task, image, cfg.task.patch_image_size)
        sample = utils.move_to_cuda(sample) if use_cuda else sample
        sample = utils.apply_to_sample(apply_half, sample) if use_fp16 else sample

        with torch.no_grad():
            result, scores = eval_step(task, generator, models, sample)
        ocr_result.append(result[0]['ocr'].replace(' ', ''))

    result = '\n'.join(ocr_result)
    return result


title = "OFA-OCR"
description = "Gradio Demo for OFA-OCR. Upload your own image or click any one of the examples, and click " \
              "\"Submit\" and then wait for the generated OCR result.  "
article = "<p style='text-align: center'><a href='https://github.com/OFA-Sys/OFA' target='_blank'>OFA Github " \
          "Repo</a></p> "
examples = [['EasyOCRLite/pics/lihe.png'], ['EasyOCRLite/pics/chinese.jpg'], ['EasyOCRLite/pics/paibian.jpeg'], \
            ['EasyOCRLite/pics/shupai.png'], ['EasyOCRLite/pics/zuowen.jpg']]
io = gr.Interface(fn=ocr, inputs=gr.inputs.Image(type='pil'), outputs=gr.outputs.Textbox(label="Caption"),
                  title=title, description=description, article=article, examples=examples,
                  allow_flagging=False, allow_screenshot=False)
io.launch(cache_examples=True)