forked from PaddlePaddle/PaddleOCR
-
Notifications
You must be signed in to change notification settings - Fork 0
/
ocr_db_crnn.cc
413 lines (346 loc) · 13.3 KB
/
ocr_db_crnn.cc
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
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http:https://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "paddle_api.h" // NOLINT
#include <chrono>
#include "cls_process.h"
#include "crnn_process.h"
#include "db_post_process.h"
using namespace paddle::lite_api; // NOLINT
using namespace std;
// fill tensor with mean and scale and trans layout: nhwc -> nchw, neon speed up
void NeonMeanScale(const float *din, float *dout, int size,
const std::vector<float> mean,
const std::vector<float> scale) {
if (mean.size() != 3 || scale.size() != 3) {
std::cerr << "[ERROR] mean or scale size must equal to 3\n";
exit(1);
}
float32x4_t vmean0 = vdupq_n_f32(mean[0]);
float32x4_t vmean1 = vdupq_n_f32(mean[1]);
float32x4_t vmean2 = vdupq_n_f32(mean[2]);
float32x4_t vscale0 = vdupq_n_f32(scale[0]);
float32x4_t vscale1 = vdupq_n_f32(scale[1]);
float32x4_t vscale2 = vdupq_n_f32(scale[2]);
float *dout_c0 = dout;
float *dout_c1 = dout + size;
float *dout_c2 = dout + size * 2;
int i = 0;
for (; i < size - 3; i += 4) {
float32x4x3_t vin3 = vld3q_f32(din);
float32x4_t vsub0 = vsubq_f32(vin3.val[0], vmean0);
float32x4_t vsub1 = vsubq_f32(vin3.val[1], vmean1);
float32x4_t vsub2 = vsubq_f32(vin3.val[2], vmean2);
float32x4_t vs0 = vmulq_f32(vsub0, vscale0);
float32x4_t vs1 = vmulq_f32(vsub1, vscale1);
float32x4_t vs2 = vmulq_f32(vsub2, vscale2);
vst1q_f32(dout_c0, vs0);
vst1q_f32(dout_c1, vs1);
vst1q_f32(dout_c2, vs2);
din += 12;
dout_c0 += 4;
dout_c1 += 4;
dout_c2 += 4;
}
for (; i < size; i++) {
*(dout_c0++) = (*(din++) - mean[0]) * scale[0];
*(dout_c1++) = (*(din++) - mean[1]) * scale[1];
*(dout_c2++) = (*(din++) - mean[2]) * scale[2];
}
}
// resize image to a size multiple of 32 which is required by the network
cv::Mat DetResizeImg(const cv::Mat img, int max_size_len,
std::vector<float> &ratio_hw) {
int w = img.cols;
int h = img.rows;
float ratio = 1.f;
int max_wh = w >= h ? w : h;
if (max_wh > max_size_len) {
if (h > w) {
ratio = static_cast<float>(max_size_len) / static_cast<float>(h);
} else {
ratio = static_cast<float>(max_size_len) / static_cast<float>(w);
}
}
int resize_h = static_cast<int>(float(h) * ratio);
int resize_w = static_cast<int>(float(w) * ratio);
if (resize_h % 32 == 0)
resize_h = resize_h;
else if (resize_h / 32 < 1 + 1e-5)
resize_h = 32;
else
resize_h = (resize_h / 32 - 1) * 32;
if (resize_w % 32 == 0)
resize_w = resize_w;
else if (resize_w / 32 < 1 + 1e-5)
resize_w = 32;
else
resize_w = (resize_w / 32 - 1) * 32;
cv::Mat resize_img;
cv::resize(img, resize_img, cv::Size(resize_w, resize_h));
ratio_hw.push_back(static_cast<float>(resize_h) / static_cast<float>(h));
ratio_hw.push_back(static_cast<float>(resize_w) / static_cast<float>(w));
return resize_img;
}
cv::Mat RunClsModel(cv::Mat img, std::shared_ptr<PaddlePredictor> predictor_cls,
const float thresh = 0.9) {
std::vector<float> mean = {0.5f, 0.5f, 0.5f};
std::vector<float> scale = {1 / 0.5f, 1 / 0.5f, 1 / 0.5f};
cv::Mat srcimg;
img.copyTo(srcimg);
cv::Mat crop_img;
img.copyTo(crop_img);
cv::Mat resize_img;
int index = 0;
float wh_ratio =
static_cast<float>(crop_img.cols) / static_cast<float>(crop_img.rows);
resize_img = ClsResizeImg(crop_img);
resize_img.convertTo(resize_img, CV_32FC3, 1 / 255.f);
const float *dimg = reinterpret_cast<const float *>(resize_img.data);
std::unique_ptr<Tensor> input_tensor0(std::move(predictor_cls->GetInput(0)));
input_tensor0->Resize({1, 3, resize_img.rows, resize_img.cols});
auto *data0 = input_tensor0->mutable_data<float>();
NeonMeanScale(dimg, data0, resize_img.rows * resize_img.cols, mean, scale);
// Run CLS predictor
predictor_cls->Run();
// Get output and run postprocess
std::unique_ptr<const Tensor> softmax_out(
std::move(predictor_cls->GetOutput(0)));
auto *softmax_scores = softmax_out->mutable_data<float>();
auto softmax_out_shape = softmax_out->shape();
float score = 0;
int label = 0;
for (int i = 0; i < softmax_out_shape[1]; i++) {
if (softmax_scores[i] > score) {
score = softmax_scores[i];
label = i;
}
}
if (label % 2 == 1 && score > thresh) {
cv::rotate(srcimg, srcimg, 1);
}
return srcimg;
}
void RunRecModel(std::vector<std::vector<std::vector<int>>> boxes, cv::Mat img,
std::shared_ptr<PaddlePredictor> predictor_crnn,
std::vector<std::string> &rec_text,
std::vector<float> &rec_text_score,
std::vector<std::string> charactor_dict,
std::shared_ptr<PaddlePredictor> predictor_cls,
int use_direction_classify) {
std::vector<float> mean = {0.5f, 0.5f, 0.5f};
std::vector<float> scale = {1 / 0.5f, 1 / 0.5f, 1 / 0.5f};
cv::Mat srcimg;
img.copyTo(srcimg);
cv::Mat crop_img;
cv::Mat resize_img;
int index = 0;
for (int i = boxes.size() - 1; i >= 0; i--) {
crop_img = GetRotateCropImage(srcimg, boxes[i]);
if (use_direction_classify >= 1) {
crop_img = RunClsModel(crop_img, predictor_cls);
}
float wh_ratio =
static_cast<float>(crop_img.cols) / static_cast<float>(crop_img.rows);
resize_img = CrnnResizeImg(crop_img, wh_ratio);
resize_img.convertTo(resize_img, CV_32FC3, 1 / 255.f);
const float *dimg = reinterpret_cast<const float *>(resize_img.data);
std::unique_ptr<Tensor> input_tensor0(
std::move(predictor_crnn->GetInput(0)));
input_tensor0->Resize({1, 3, resize_img.rows, resize_img.cols});
auto *data0 = input_tensor0->mutable_data<float>();
NeonMeanScale(dimg, data0, resize_img.rows * resize_img.cols, mean, scale);
//// Run CRNN predictor
predictor_crnn->Run();
// Get output and run postprocess
std::unique_ptr<const Tensor> output_tensor0(
std::move(predictor_crnn->GetOutput(0)));
auto *predict_batch = output_tensor0->data<float>();
auto predict_shape = output_tensor0->shape();
// ctc decode
std::string str_res;
int argmax_idx;
int last_index = 0;
float score = 0.f;
int count = 0;
float max_value = 0.0f;
for (int n = 0; n < predict_shape[1]; n++) {
argmax_idx = int(Argmax(&predict_batch[n * predict_shape[2]],
&predict_batch[(n + 1) * predict_shape[2]]));
max_value =
float(*std::max_element(&predict_batch[n * predict_shape[2]],
&predict_batch[(n + 1) * predict_shape[2]]));
if (argmax_idx > 0 && (!(n > 0 && argmax_idx == last_index))) {
score += max_value;
count += 1;
str_res += charactor_dict[argmax_idx];
}
last_index = argmax_idx;
}
score /= count;
rec_text.push_back(str_res);
rec_text_score.push_back(score);
}
}
std::vector<std::vector<std::vector<int>>>
RunDetModel(std::shared_ptr<PaddlePredictor> predictor, cv::Mat img,
std::map<std::string, double> Config) {
// Read img
int max_side_len = int(Config["max_side_len"]);
int det_db_use_dilate = int(Config["det_db_use_dilate"]);
cv::Mat srcimg;
img.copyTo(srcimg);
std::vector<float> ratio_hw;
img = DetResizeImg(img, max_side_len, ratio_hw);
cv::Mat img_fp;
img.convertTo(img_fp, CV_32FC3, 1.0 / 255.f);
// Prepare input data from image
std::unique_ptr<Tensor> input_tensor0(std::move(predictor->GetInput(0)));
input_tensor0->Resize({1, 3, img_fp.rows, img_fp.cols});
auto *data0 = input_tensor0->mutable_data<float>();
std::vector<float> mean = {0.485f, 0.456f, 0.406f};
std::vector<float> scale = {1 / 0.229f, 1 / 0.224f, 1 / 0.225f};
const float *dimg = reinterpret_cast<const float *>(img_fp.data);
NeonMeanScale(dimg, data0, img_fp.rows * img_fp.cols, mean, scale);
// Run predictor
predictor->Run();
// Get output and post process
std::unique_ptr<const Tensor> output_tensor(
std::move(predictor->GetOutput(0)));
auto *outptr = output_tensor->data<float>();
auto shape_out = output_tensor->shape();
// Save output
float pred[shape_out[2] * shape_out[3]];
unsigned char cbuf[shape_out[2] * shape_out[3]];
for (int i = 0; i < int(shape_out[2] * shape_out[3]); i++) {
pred[i] = static_cast<float>(outptr[i]);
cbuf[i] = static_cast<unsigned char>((outptr[i]) * 255);
}
cv::Mat cbuf_map(shape_out[2], shape_out[3], CV_8UC1,
reinterpret_cast<unsigned char *>(cbuf));
cv::Mat pred_map(shape_out[2], shape_out[3], CV_32F,
reinterpret_cast<float *>(pred));
const double threshold = double(Config["det_db_thresh"]) * 255;
const double max_value = 255;
cv::Mat bit_map;
cv::threshold(cbuf_map, bit_map, threshold, max_value, cv::THRESH_BINARY);
if (det_db_use_dilate == 1) {
cv::Mat dilation_map;
cv::Mat dila_ele =
cv::getStructuringElement(cv::MORPH_RECT, cv::Size(2, 2));
cv::dilate(bit_map, dilation_map, dila_ele);
bit_map = dilation_map;
}
auto boxes = BoxesFromBitmap(pred_map, bit_map, Config);
std::vector<std::vector<std::vector<int>>> filter_boxes =
FilterTagDetRes(boxes, ratio_hw[0], ratio_hw[1], srcimg);
return filter_boxes;
}
std::shared_ptr<PaddlePredictor> loadModel(std::string model_file) {
MobileConfig config;
config.set_model_from_file(model_file);
std::shared_ptr<PaddlePredictor> predictor =
CreatePaddlePredictor<MobileConfig>(config);
return predictor;
}
cv::Mat Visualization(cv::Mat srcimg,
std::vector<std::vector<std::vector<int>>> boxes) {
cv::Point rook_points[boxes.size()][4];
for (int n = 0; n < boxes.size(); n++) {
for (int m = 0; m < boxes[0].size(); m++) {
rook_points[n][m] = cv::Point(static_cast<int>(boxes[n][m][0]),
static_cast<int>(boxes[n][m][1]));
}
}
cv::Mat img_vis;
srcimg.copyTo(img_vis);
for (int n = 0; n < boxes.size(); n++) {
const cv::Point *ppt[1] = {rook_points[n]};
int npt[] = {4};
cv::polylines(img_vis, ppt, npt, 1, 1, CV_RGB(0, 255, 0), 2, 8, 0);
}
cv::imwrite("./vis.jpg", img_vis);
std::cout << "The detection visualized image saved in ./vis.jpg" << std::endl;
return img_vis;
}
std::vector<std::string> split(const std::string &str,
const std::string &delim) {
std::vector<std::string> res;
if ("" == str)
return res;
char *strs = new char[str.length() + 1];
std::strcpy(strs, str.c_str());
char *d = new char[delim.length() + 1];
std::strcpy(d, delim.c_str());
char *p = std::strtok(strs, d);
while (p) {
string s = p;
res.push_back(s);
p = std::strtok(NULL, d);
}
return res;
}
std::map<std::string, double> LoadConfigTxt(std::string config_path) {
auto config = ReadDict(config_path);
std::map<std::string, double> dict;
for (int i = 0; i < config.size(); i++) {
std::vector<std::string> res = split(config[i], " ");
dict[res[0]] = stod(res[1]);
}
return dict;
}
int main(int argc, char **argv) {
if (argc < 5) {
std::cerr << "[ERROR] usage: " << argv[0]
<< " det_model_file cls_model_file rec_model_file image_path "
"charactor_dict\n";
exit(1);
}
std::string det_model_file = argv[1];
std::string rec_model_file = argv[2];
std::string cls_model_file = argv[3];
std::string img_path = argv[4];
std::string dict_path = argv[5];
//// load config from txt file
auto Config = LoadConfigTxt("./config.txt");
int use_direction_classify = int(Config["use_direction_classify"]);
auto start = std::chrono::system_clock::now();
auto det_predictor = loadModel(det_model_file);
auto rec_predictor = loadModel(rec_model_file);
auto cls_predictor = loadModel(cls_model_file);
auto charactor_dict = ReadDict(dict_path);
charactor_dict.insert(charactor_dict.begin(), "#"); // blank char for ctc
charactor_dict.push_back(" ");
cv::Mat srcimg = cv::imread(img_path, cv::IMREAD_COLOR);
auto boxes = RunDetModel(det_predictor, srcimg, Config);
std::vector<std::string> rec_text;
std::vector<float> rec_text_score;
RunRecModel(boxes, srcimg, rec_predictor, rec_text, rec_text_score,
charactor_dict, cls_predictor, use_direction_classify);
auto end = std::chrono::system_clock::now();
auto duration =
std::chrono::duration_cast<std::chrono::microseconds>(end - start);
//// visualization
auto img_vis = Visualization(srcimg, boxes);
//// print recognized text
for (int i = 0; i < rec_text.size(); i++) {
std::cout << i << "\t" << rec_text[i] << "\t" << rec_text_score[i]
<< std::endl;
}
std::cout << "花费了"
<< double(duration.count()) *
std::chrono::microseconds::period::num /
std::chrono::microseconds::period::den
<< "秒" << std::endl;
return 0;
}