infer.py 30.0 KB
Newer Older
Q
qingqing01 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
# 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://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.

import os
import yaml
G
Guanghua Yu 已提交
17
import glob
Q
qingqing01 已提交
18 19 20 21
from functools import reduce

import cv2
import numpy as np
C
cnn 已提交
22
import math
Q
qingqing01 已提交
23 24 25 26
import paddle
from paddle.inference import Config
from paddle.inference import create_predictor

27
from benchmark_utils import PaddleInferBenchmark
28
from picodet_postprocess import PicoDetPostProcess
W
wangguanzhong 已提交
29
from preprocess import preprocess, Resize, NormalizeImage, Permute, PadStride, LetterBoxResize, WarpAffine
G
Guanghua Yu 已提交
30
from visualize import visualize_box_mask
31
from utils import argsparser, Timer, get_current_memory_mb
G
Guanghua Yu 已提交
32

Q
qingqing01 已提交
33 34 35 36 37
# Global dictionary
SUPPORT_MODELS = {
    'YOLO',
    'RCNN',
    'SSD',
38
    'Face',
F
Feng Ni 已提交
39
    'FCOS',
G
Guanghua Yu 已提交
40
    'SOLOv2',
F
Feng Ni 已提交
41
    'TTFNet',
C
cnn 已提交
42
    'S2ANet',
G
George Ni 已提交
43 44 45
    'JDE',
    'FairMOT',
    'DeepSORT',
G
Guanghua Yu 已提交
46 47
    'GFL',
    'PicoDet',
W
wangguanzhong 已提交
48
    'CenterNet',
Q
qingqing01 已提交
49 50 51 52 53 54
}


class Detector(object):
    """
    Args:
55
        pred_config (object): config of model, defined by `Config(model_dir)`
Q
qingqing01 已提交
56
        model_dir (str): root path of model.pdiparams, model.pdmodel and infer_cfg.yml
G
Guanghua Yu 已提交
57
        device (str): Choose the device you want to run, it can be: CPU/GPU/XPU, default is CPU
Q
qingqing01 已提交
58
        run_mode (str): mode of running(fluid/trt_fp32/trt_fp16)
59
        batch_size (int): size of pre batch in inference
60 61 62
        trt_min_shape (int): min shape for dynamic shape in trt
        trt_max_shape (int): max shape for dynamic shape in trt
        trt_opt_shape (int): opt shape for dynamic shape in trt
63 64 65 66
        trt_calib_mode (bool): If the model is produced by TRT offline quantitative
            calibration, trt_calib_mode need to set True
        cpu_threads (int): cpu threads
        enable_mkldnn (bool): whether to open MKLDNN
Q
qingqing01 已提交
67 68 69 70 71
    """

    def __init__(self,
                 pred_config,
                 model_dir,
G
Guanghua Yu 已提交
72
                 device='CPU',
Q
qingqing01 已提交
73
                 run_mode='fluid',
74
                 batch_size=1,
75 76 77
                 trt_min_shape=1,
                 trt_max_shape=1280,
                 trt_opt_shape=640,
78 79 80
                 trt_calib_mode=False,
                 cpu_threads=1,
                 enable_mkldnn=False):
Q
qingqing01 已提交
81
        self.pred_config = pred_config
82
        self.predictor, self.config = load_predictor(
Q
qingqing01 已提交
83 84
            model_dir,
            run_mode=run_mode,
85
            batch_size=batch_size,
Q
qingqing01 已提交
86
            min_subgraph_size=self.pred_config.min_subgraph_size,
G
Guanghua Yu 已提交
87
            device=device,
88
            use_dynamic_shape=self.pred_config.use_dynamic_shape,
89 90
            trt_min_shape=trt_min_shape,
            trt_max_shape=trt_max_shape,
G
Guanghua Yu 已提交
91
            trt_opt_shape=trt_opt_shape,
92 93 94
            trt_calib_mode=trt_calib_mode,
            cpu_threads=cpu_threads,
            enable_mkldnn=enable_mkldnn)
G
Guanghua Yu 已提交
95 96
        self.det_times = Timer()
        self.cpu_mem, self.gpu_mem, self.gpu_util = 0, 0, 0
Q
qingqing01 已提交
97

C
cnn 已提交
98
    def preprocess(self, image_list):
Q
qingqing01 已提交
99 100 101 102 103
        preprocess_ops = []
        for op_info in self.pred_config.preprocess_infos:
            new_op_info = op_info.copy()
            op_type = new_op_info.pop('type')
            preprocess_ops.append(eval(op_type)(**new_op_info))
C
cnn 已提交
104 105 106 107

        input_im_lst = []
        input_im_info_lst = []
        for im_path in image_list:
108
            im, im_info = preprocess(im_path, preprocess_ops)
C
cnn 已提交
109 110 111
            input_im_lst.append(im)
            input_im_info_lst.append(im_info)
        inputs = create_inputs(input_im_lst, input_im_info_lst)
Q
qingqing01 已提交
112 113
        return inputs

C
cnn 已提交
114 115 116 117 118 119
    def postprocess(self,
                    np_boxes,
                    np_masks,
                    inputs,
                    np_boxes_num,
                    threshold=0.5):
Q
qingqing01 已提交
120 121 122
        # postprocess output of predictor
        results = {}
        results['boxes'] = np_boxes
C
cnn 已提交
123
        results['boxes_num'] = np_boxes_num
Q
qingqing01 已提交
124 125 126 127
        if np_masks is not None:
            results['masks'] = np_masks
        return results

C
cnn 已提交
128
    def predict(self, image_list, threshold=0.5, warmup=0, repeats=1):
Q
qingqing01 已提交
129 130
        '''
        Args:
131
            image_list (list): list of image
Q
qingqing01 已提交
132 133 134 135 136
            threshold (float): threshold of predicted box' score
        Returns:
            results (dict): include 'boxes': np.ndarray: shape:[N,6], N: number of box,
                            matix element:[class, score, x_min, y_min, x_max, y_max]
                            MaskRCNN's results include 'masks': np.ndarray:
G
Guanghua Yu 已提交
137
                            shape: [N, im_h, im_w]
Q
qingqing01 已提交
138
        '''
139
        self.det_times.preprocess_time_s.start()
C
cnn 已提交
140
        inputs = self.preprocess(image_list)
141
        self.det_times.preprocess_time_s.end()
Q
qingqing01 已提交
142 143 144 145 146 147 148 149 150 151
        np_boxes, np_masks = None, None
        input_names = self.predictor.get_input_names()
        for i in range(len(input_names)):
            input_tensor = self.predictor.get_input_handle(input_names[i])
            input_tensor.copy_from_cpu(inputs[input_names[i]])
        for i in range(warmup):
            self.predictor.run()
            output_names = self.predictor.get_output_names()
            boxes_tensor = self.predictor.get_output_handle(output_names[0])
            np_boxes = boxes_tensor.copy_to_cpu()
G
Guanghua Yu 已提交
152
            if self.pred_config.mask:
Q
qingqing01 已提交
153 154 155
                masks_tensor = self.predictor.get_output_handle(output_names[2])
                np_masks = masks_tensor.copy_to_cpu()

156
        self.det_times.inference_time_s.start()
Q
qingqing01 已提交
157 158 159 160 161
        for i in range(repeats):
            self.predictor.run()
            output_names = self.predictor.get_output_names()
            boxes_tensor = self.predictor.get_output_handle(output_names[0])
            np_boxes = boxes_tensor.copy_to_cpu()
C
cnn 已提交
162 163
            boxes_num = self.predictor.get_output_handle(output_names[1])
            np_boxes_num = boxes_num.copy_to_cpu()
G
Guanghua Yu 已提交
164
            if self.pred_config.mask:
Q
qingqing01 已提交
165 166
                masks_tensor = self.predictor.get_output_handle(output_names[2])
                np_masks = masks_tensor.copy_to_cpu()
167
        self.det_times.inference_time_s.end(repeats=repeats)
Q
qingqing01 已提交
168

169
        self.det_times.postprocess_time_s.start()
Q
qingqing01 已提交
170
        results = []
G
Guanghua Yu 已提交
171 172
        if reduce(lambda x, y: x * y, np_boxes.shape) < 6:
            print('[WARNNING] No object detected.')
173
            results = {'boxes': np.zeros([0, 6]), 'boxes_num': [0]}
G
Guanghua Yu 已提交
174 175
        else:
            results = self.postprocess(
C
cnn 已提交
176
                np_boxes, np_masks, inputs, np_boxes_num, threshold=threshold)
177
        self.det_times.postprocess_time_s.end()
C
cnn 已提交
178
        self.det_times.img_num += len(image_list)
Q
qingqing01 已提交
179 180
        return results

W
wangguanzhong 已提交
181 182 183
    def get_timer(self):
        return self.det_times

Q
qingqing01 已提交
184

G
Guanghua Yu 已提交
185 186 187 188 189
class DetectorSOLOv2(Detector):
    """
    Args:
        config (object): config of model, defined by `Config(model_dir)`
        model_dir (str): root path of model.pdiparams, model.pdmodel and infer_cfg.yml
G
Guanghua Yu 已提交
190
        device (str): Choose the device you want to run, it can be: CPU/GPU/XPU, default is CPU
G
Guanghua Yu 已提交
191
        run_mode (str): mode of running(fluid/trt_fp32/trt_fp16)
192
        batch_size (int): size of pre batch in inference
193 194 195
        trt_min_shape (int): min shape for dynamic shape in trt
        trt_max_shape (int): max shape for dynamic shape in trt
        trt_opt_shape (int): opt shape for dynamic shape in trt
196 197 198 199
        trt_calib_mode (bool): If the model is produced by TRT offline quantitative
            calibration, trt_calib_mode need to set True
        cpu_threads (int): cpu threads
        enable_mkldnn (bool): whether to open MKLDNN 
G
Guanghua Yu 已提交
200 201 202 203 204
    """

    def __init__(self,
                 pred_config,
                 model_dir,
G
Guanghua Yu 已提交
205
                 device='CPU',
G
Guanghua Yu 已提交
206
                 run_mode='fluid',
207
                 batch_size=1,
208 209 210
                 trt_min_shape=1,
                 trt_max_shape=1280,
                 trt_opt_shape=640,
211 212 213
                 trt_calib_mode=False,
                 cpu_threads=1,
                 enable_mkldnn=False):
G
Guanghua Yu 已提交
214
        self.pred_config = pred_config
215
        self.predictor, self.config = load_predictor(
G
Guanghua Yu 已提交
216 217
            model_dir,
            run_mode=run_mode,
218
            batch_size=batch_size,
G
Guanghua Yu 已提交
219
            min_subgraph_size=self.pred_config.min_subgraph_size,
G
Guanghua Yu 已提交
220
            device=device,
221
            use_dynamic_shape=self.pred_config.use_dynamic_shape,
222 223
            trt_min_shape=trt_min_shape,
            trt_max_shape=trt_max_shape,
G
Guanghua Yu 已提交
224
            trt_opt_shape=trt_opt_shape,
225 226 227
            trt_calib_mode=trt_calib_mode,
            cpu_threads=cpu_threads,
            enable_mkldnn=enable_mkldnn)
G
Guanghua Yu 已提交
228
        self.det_times = Timer()
229
        self.cpu_mem, self.gpu_mem, self.gpu_util = 0, 0, 0
G
Guanghua Yu 已提交
230 231

    def predict(self, image, threshold=0.5, warmup=0, repeats=1):
G
Guanghua Yu 已提交
232 233 234 235 236
        '''
        Args:
            image (str/np.ndarray): path of image/ np.ndarray read by cv2
            threshold (float): threshold of predicted box' score
        Returns:
G
Guanghua Yu 已提交
237 238 239
            results (dict): 'segm': np.ndarray,shape:[N, im_h, im_w]
                            'cate_label': label of segm, shape:[N]
                            'cate_score': confidence score of segm, shape:[N]
G
Guanghua Yu 已提交
240
        '''
241
        self.det_times.preprocess_time_s.start()
G
Guanghua Yu 已提交
242
        inputs = self.preprocess(image)
243
        self.det_times.preprocess_time_s.end()
G
Guanghua Yu 已提交
244 245 246 247 248 249 250 251
        np_label, np_score, np_segms = None, None, None
        input_names = self.predictor.get_input_names()
        for i in range(len(input_names)):
            input_tensor = self.predictor.get_input_handle(input_names[i])
            input_tensor.copy_from_cpu(inputs[input_names[i]])
        for i in range(warmup):
            self.predictor.run()
            output_names = self.predictor.get_output_names()
W
wangguanzhong 已提交
252 253
            np_boxes_num = self.predictor.get_output_handle(output_names[
                0]).copy_to_cpu()
G
Guanghua Yu 已提交
254 255
            np_label = self.predictor.get_output_handle(output_names[
                1]).copy_to_cpu()
G
Guanghua Yu 已提交
256
            np_score = self.predictor.get_output_handle(output_names[
G
Guanghua Yu 已提交
257
                2]).copy_to_cpu()
G
Guanghua Yu 已提交
258 259
            np_segms = self.predictor.get_output_handle(output_names[
                3]).copy_to_cpu()
260
        self.det_times.inference_time_s.start()
G
Guanghua Yu 已提交
261 262 263
        for i in range(repeats):
            self.predictor.run()
            output_names = self.predictor.get_output_names()
W
wangguanzhong 已提交
264 265
            np_boxes_num = self.predictor.get_output_handle(output_names[
                0]).copy_to_cpu()
G
Guanghua Yu 已提交
266 267
            np_label = self.predictor.get_output_handle(output_names[
                1]).copy_to_cpu()
G
Guanghua Yu 已提交
268
            np_score = self.predictor.get_output_handle(output_names[
G
Guanghua Yu 已提交
269
                2]).copy_to_cpu()
G
Guanghua Yu 已提交
270 271
            np_segms = self.predictor.get_output_handle(output_names[
                3]).copy_to_cpu()
272
        self.det_times.inference_time_s.end(repeats=repeats)
G
Guanghua Yu 已提交
273
        self.det_times.img_num += 1
G
Guanghua Yu 已提交
274

W
wangguanzhong 已提交
275 276 277 278 279
        return dict(
            segm=np_segms,
            label=np_label,
            score=np_score,
            boxes_num=np_boxes_num)
G
Guanghua Yu 已提交
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
class DetectorPicoDet(Detector):
    """
    Args:
        config (object): config of model, defined by `Config(model_dir)`
        model_dir (str): root path of model.pdiparams, model.pdmodel and infer_cfg.yml
        device (str): Choose the device you want to run, it can be: CPU/GPU/XPU, default is CPU
        run_mode (str): mode of running(fluid/trt_fp32/trt_fp16)
        batch_size (int): size of pre batch in inference
        trt_min_shape (int): min shape for dynamic shape in trt
        trt_max_shape (int): max shape for dynamic shape in trt
        trt_opt_shape (int): opt shape for dynamic shape in trt
        trt_calib_mode (bool): If the model is produced by TRT offline quantitative
            calibration, trt_calib_mode need to set True
        cpu_threads (int): cpu threads
        enable_mkldnn (bool): whether to open MKLDNN 
    """

    def __init__(self,
                 pred_config,
                 model_dir,
                 device='CPU',
                 run_mode='fluid',
                 batch_size=1,
                 trt_min_shape=1,
                 trt_max_shape=1280,
                 trt_opt_shape=640,
                 trt_calib_mode=False,
                 cpu_threads=1,
                 enable_mkldnn=False):
        self.pred_config = pred_config
        self.predictor, self.config = load_predictor(
            model_dir,
            run_mode=run_mode,
            batch_size=batch_size,
            min_subgraph_size=self.pred_config.min_subgraph_size,
            device=device,
            use_dynamic_shape=self.pred_config.use_dynamic_shape,
            trt_min_shape=trt_min_shape,
            trt_max_shape=trt_max_shape,
            trt_opt_shape=trt_opt_shape,
            trt_calib_mode=trt_calib_mode,
            cpu_threads=cpu_threads,
            enable_mkldnn=enable_mkldnn)
        self.det_times = Timer()
        self.cpu_mem, self.gpu_mem, self.gpu_util = 0, 0, 0

    def predict(self, image, threshold=0.5, warmup=0, repeats=1):
        '''
        Args:
            image (str/np.ndarray): path of image/ np.ndarray read by cv2
            threshold (float): threshold of predicted box' score
        Returns:
            results (dict): include 'boxes': np.ndarray: shape:[N,6], N: number of box,
                            matix element:[class, score, x_min, y_min, x_max, y_max]
        '''
        self.det_times.preprocess_time_s.start()
        inputs = self.preprocess(image)
        self.det_times.preprocess_time_s.end()
        input_names = self.predictor.get_input_names()
        for i in range(len(input_names)):
            input_tensor = self.predictor.get_input_handle(input_names[i])
            input_tensor.copy_from_cpu(inputs[input_names[i]])
        np_score_list, np_boxes_list = [], []
        for i in range(warmup):
            self.predictor.run()
            np_score_list.clear()
            np_boxes_list.clear()
            output_names = self.predictor.get_output_names()
            num_outs = int(len(output_names) / 2)
            for out_idx in range(num_outs):
                np_score_list.append(
                    self.predictor.get_output_handle(output_names[out_idx])
                    .copy_to_cpu())
                np_boxes_list.append(
                    self.predictor.get_output_handle(output_names[
                        out_idx + num_outs]).copy_to_cpu())

        self.det_times.inference_time_s.start()
        for i in range(repeats):
            self.predictor.run()
            np_score_list.clear()
            np_boxes_list.clear()
            output_names = self.predictor.get_output_names()
            num_outs = int(len(output_names) / 2)
            for out_idx in range(num_outs):
                np_score_list.append(
                    self.predictor.get_output_handle(output_names[out_idx])
                    .copy_to_cpu())
                np_boxes_list.append(
                    self.predictor.get_output_handle(output_names[
                        out_idx + num_outs]).copy_to_cpu())
        self.det_times.inference_time_s.end(repeats=repeats)
        self.det_times.img_num += 1
        self.det_times.postprocess_time_s.start()
        self.postprocess = PicoDetPostProcess(
            inputs['image'].shape[2:],
            inputs['im_shape'],
            inputs['scale_factor'],
            strides=self.pred_config.fpn_stride,
            nms_threshold=self.pred_config.nms['nms_threshold'])
        np_boxes, np_boxes_num = self.postprocess(np_score_list, np_boxes_list)
        self.det_times.postprocess_time_s.end()
        return dict(boxes=np_boxes, boxes_num=np_boxes_num)


C
cnn 已提交
387
def create_inputs(imgs, im_info):
Q
qingqing01 已提交
388 389
    """generate input for different model type
    Args:
W
wangguanzhong 已提交
390 391
        imgs (list(numpy)): list of images (np.ndarray)
        im_info (list(dict)): list of image info
Q
qingqing01 已提交
392 393 394 395 396
    Returns:
        inputs (dict): input of model
    """
    inputs = {}

C
cnn 已提交
397 398
    im_shape = []
    scale_factor = []
399 400 401 402 403 404 405 406
    if len(imgs) == 1:
        inputs['image'] = np.array((imgs[0], )).astype('float32')
        inputs['im_shape'] = np.array(
            (im_info[0]['im_shape'], )).astype('float32')
        inputs['scale_factor'] = np.array(
            (im_info[0]['scale_factor'], )).astype('float32')
        return inputs

C
cnn 已提交
407 408 409 410
    for e in im_info:
        im_shape.append(np.array((e['im_shape'], )).astype('float32'))
        scale_factor.append(np.array((e['scale_factor'], )).astype('float32'))

C
cnn 已提交
411 412
    inputs['im_shape'] = np.concatenate(im_shape, axis=0)
    inputs['scale_factor'] = np.concatenate(scale_factor, axis=0)
C
cnn 已提交
413 414 415 416 417 418 419 420 421 422 423 424

    imgs_shape = [[e.shape[1], e.shape[2]] for e in imgs]
    max_shape_h = max([e[0] for e in imgs_shape])
    max_shape_w = max([e[1] for e in imgs_shape])
    padding_imgs = []
    for img in imgs:
        im_c, im_h, im_w = img.shape[:]
        padding_im = np.zeros(
            (im_c, max_shape_h, max_shape_w), dtype=np.float32)
        padding_im[:, :im_h, :im_w] = img
        padding_imgs.append(padding_im)
    inputs['image'] = np.stack(padding_imgs, axis=0)
Q
qingqing01 已提交
425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443
    return inputs


class PredictConfig():
    """set config of preprocess, postprocess and visualize
    Args:
        model_dir (str): root path of model.yml
    """

    def __init__(self, model_dir):
        # parsing Yaml config for Preprocess
        deploy_file = os.path.join(model_dir, 'infer_cfg.yml')
        with open(deploy_file) as f:
            yml_conf = yaml.safe_load(f)
        self.check_model(yml_conf)
        self.arch = yml_conf['arch']
        self.preprocess_infos = yml_conf['Preprocess']
        self.min_subgraph_size = yml_conf['min_subgraph_size']
        self.labels = yml_conf['label_list']
G
Guanghua Yu 已提交
444
        self.mask = False
445
        self.use_dynamic_shape = yml_conf['use_dynamic_shape']
G
Guanghua Yu 已提交
446 447
        if 'mask' in yml_conf:
            self.mask = yml_conf['mask']
448 449 450
        self.tracker = None
        if 'tracker' in yml_conf:
            self.tracker = yml_conf['tracker']
451 452 453 454
        if 'NMS' in yml_conf:
            self.nms = yml_conf['NMS']
        if 'fpn_stride' in yml_conf:
            self.fpn_stride = yml_conf['fpn_stride']
Q
qingqing01 已提交
455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479
        self.print_config()

    def check_model(self, yml_conf):
        """
        Raises:
            ValueError: loaded model not in supported model type 
        """
        for support_model in SUPPORT_MODELS:
            if support_model in yml_conf['arch']:
                return True
        raise ValueError("Unsupported arch: {}, expect {}".format(yml_conf[
            'arch'], SUPPORT_MODELS))

    def print_config(self):
        print('-----------  Model Configuration -----------')
        print('%s: %s' % ('Model Arch', self.arch))
        print('%s: ' % ('Transform Order'))
        for op_info in self.preprocess_infos:
            print('--%s: %s' % ('transform op', op_info['type']))
        print('--------------------------------------------')


def load_predictor(model_dir,
                   run_mode='fluid',
                   batch_size=1,
G
Guanghua Yu 已提交
480
                   device='CPU',
481 482 483 484
                   min_subgraph_size=3,
                   use_dynamic_shape=False,
                   trt_min_shape=1,
                   trt_max_shape=1280,
G
Guanghua Yu 已提交
485
                   trt_opt_shape=640,
486 487 488
                   trt_calib_mode=False,
                   cpu_threads=1,
                   enable_mkldnn=False):
Q
qingqing01 已提交
489 490 491
    """set AnalysisConfig, generate AnalysisPredictor
    Args:
        model_dir (str): root path of __model__ and __params__
G
Guanghua Yu 已提交
492
        device (str): Choose the device you want to run, it can be: CPU/GPU/XPU, default is CPU
493
        run_mode (str): mode of running(fluid/trt_fp32/trt_fp16/trt_int8)
494 495 496 497
        use_dynamic_shape (bool): use dynamic shape or not
        trt_min_shape (int): min shape for dynamic shape in trt
        trt_max_shape (int): max shape for dynamic shape in trt
        trt_opt_shape (int): opt shape for dynamic shape in trt
G
Guanghua Yu 已提交
498 499
        trt_calib_mode (bool): If the model is produced by TRT offline quantitative
            calibration, trt_calib_mode need to set True
Q
qingqing01 已提交
500 501 502
    Returns:
        predictor (PaddlePredictor): AnalysisPredictor
    Raises:
G
Guanghua Yu 已提交
503
        ValueError: predict by TensorRT need device == 'GPU'.
Q
qingqing01 已提交
504
    """
G
Guanghua Yu 已提交
505
    if device != 'GPU' and run_mode != 'fluid':
Q
qingqing01 已提交
506
        raise ValueError(
G
Guanghua Yu 已提交
507 508
            "Predict by TensorRT mode: {}, expect device=='GPU', but device == {}"
            .format(run_mode, device))
Q
qingqing01 已提交
509 510 511
    config = Config(
        os.path.join(model_dir, 'model.pdmodel'),
        os.path.join(model_dir, 'model.pdiparams'))
G
Guanghua Yu 已提交
512
    if device == 'GPU':
Q
qingqing01 已提交
513 514 515
        # initial GPU memory(M), device ID
        config.enable_use_gpu(200, 0)
        # optimize graph and fuse op
516
        config.switch_ir_optim(True)
G
Guanghua Yu 已提交
517 518
    elif device == 'XPU':
        config.enable_xpu(10 * 1024 * 1024)
Q
qingqing01 已提交
519 520
    else:
        config.disable_gpu()
521 522
        config.set_cpu_math_library_num_threads(cpu_threads)
        if enable_mkldnn:
G
Guanghua Yu 已提交
523 524 525 526 527 528 529 530 531
            try:
                # cache 10 different shapes for mkldnn to avoid memory leak
                config.set_mkldnn_cache_capacity(10)
                config.enable_mkldnn()
            except Exception as e:
                print(
                    "The current environment does not support `mkldnn`, so disable mkldnn."
                )
                pass
Q
qingqing01 已提交
532

G
Guanghua Yu 已提交
533 534 535 536 537
    precision_map = {
        'trt_int8': Config.Precision.Int8,
        'trt_fp32': Config.Precision.Float32,
        'trt_fp16': Config.Precision.Half
    }
Q
qingqing01 已提交
538 539
    if run_mode in precision_map.keys():
        config.enable_tensorrt_engine(
540
            workspace_size=1 << 25,
Q
qingqing01 已提交
541 542 543 544
            max_batch_size=batch_size,
            min_subgraph_size=min_subgraph_size,
            precision_mode=precision_map[run_mode],
            use_static=False,
G
Guanghua Yu 已提交
545
            use_calib_mode=trt_calib_mode)
546 547

        if use_dynamic_shape:
548 549 550 551 552 553 554 555 556
            min_input_shape = {
                'image': [batch_size, 3, trt_min_shape, trt_min_shape]
            }
            max_input_shape = {
                'image': [batch_size, 3, trt_max_shape, trt_max_shape]
            }
            opt_input_shape = {
                'image': [batch_size, 3, trt_opt_shape, trt_opt_shape]
            }
557 558 559
            config.set_trt_dynamic_shape_info(min_input_shape, max_input_shape,
                                              opt_input_shape)
            print('trt set dynamic shape done!')
Q
qingqing01 已提交
560 561 562 563 564 565 566 567

    # disable print log when predict
    config.disable_glog_info()
    # enable shared memory
    config.enable_memory_optim()
    # disable feed, fetch OP, needed by zero_copy_run
    config.switch_use_feed_fetch_ops(False)
    predictor = create_predictor(config)
568
    return predictor, config
Q
qingqing01 已提交
569 570


G
Guanghua Yu 已提交
571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601
def get_test_images(infer_dir, infer_img):
    """
    Get image path list in TEST mode
    """
    assert infer_img is not None or infer_dir is not None, \
        "--infer_img or --infer_dir should be set"
    assert infer_img is None or os.path.isfile(infer_img), \
            "{} is not a file".format(infer_img)
    assert infer_dir is None or os.path.isdir(infer_dir), \
            "{} is not a directory".format(infer_dir)

    # infer_img has a higher priority
    if infer_img and os.path.isfile(infer_img):
        return [infer_img]

    images = set()
    infer_dir = os.path.abspath(infer_dir)
    assert os.path.isdir(infer_dir), \
        "infer_dir {} is not a directory".format(infer_dir)
    exts = ['jpg', 'jpeg', 'png', 'bmp']
    exts += [ext.upper() for ext in exts]
    for ext in exts:
        images.update(glob.glob('{}/*.{}'.format(infer_dir, ext)))
    images = list(images)

    assert len(images) > 0, "no image found in {}".format(infer_dir)
    print("Found {} inference images in total.".format(len(images)))

    return images


C
cnn 已提交
602
def visualize(image_list, results, labels, output_dir='output/', threshold=0.5):
Q
qingqing01 已提交
603
    # visualize the predict result
C
cnn 已提交
604 605 606 607 608 609 610 611 612 613 614 615 616
    start_idx = 0
    for idx, image_file in enumerate(image_list):
        im_bboxes_num = results['boxes_num'][idx]
        im_results = {}
        if 'boxes' in results:
            im_results['boxes'] = results['boxes'][start_idx:start_idx +
                                                   im_bboxes_num, :]
        if 'masks' in results:
            im_results['masks'] = results['masks'][start_idx:start_idx +
                                                   im_bboxes_num, :]
        if 'segm' in results:
            im_results['segm'] = results['segm'][start_idx:start_idx +
                                                 im_bboxes_num, :]
W
wangguanzhong 已提交
617 618 619 620 621 622 623
        if 'label' in results:
            im_results['label'] = results['label'][start_idx:start_idx +
                                                   im_bboxes_num]
        if 'score' in results:
            im_results['score'] = results['score'][start_idx:start_idx +
                                                   im_bboxes_num]

C
cnn 已提交
624 625 626 627 628 629 630 631 632
        start_idx += im_bboxes_num
        im = visualize_box_mask(
            image_file, im_results, labels, threshold=threshold)
        img_name = os.path.split(image_file)[-1]
        if not os.path.exists(output_dir):
            os.makedirs(output_dir)
        out_path = os.path.join(output_dir, img_name)
        im.save(out_path, quality=95)
        print("save result to: " + out_path)
Q
qingqing01 已提交
633 634 635 636 637 638 639 640 641


def print_arguments(args):
    print('-----------  Running Arguments -----------')
    for arg, value in sorted(vars(args).items()):
        print('%s: %s' % (arg, value))
    print('------------------------------------------')


C
cnn 已提交
642 643 644 645 646 647
def predict_image(detector, image_list, batch_size=1):
    batch_loop_cnt = math.ceil(float(len(image_list)) / batch_size)
    for i in range(batch_loop_cnt):
        start_index = i * batch_size
        end_index = min((i + 1) * batch_size, len(image_list))
        batch_image_list = image_list[start_index:end_index]
G
Guanghua Yu 已提交
648
        if FLAGS.run_benchmark:
C
cnn 已提交
649 650
            detector.predict(
                batch_image_list, FLAGS.threshold, warmup=10, repeats=10)
G
Guanghua Yu 已提交
651 652 653 654
            cm, gm, gu = get_current_memory_mb()
            detector.cpu_mem += cm
            detector.gpu_mem += gm
            detector.gpu_util += gu
C
cnn 已提交
655
            print('Test iter {}'.format(i))
G
Guanghua Yu 已提交
656
        else:
C
cnn 已提交
657
            results = detector.predict(batch_image_list, FLAGS.threshold)
G
Guanghua Yu 已提交
658
            visualize(
C
cnn 已提交
659
                batch_image_list,
G
Guanghua Yu 已提交
660 661 662 663
                results,
                detector.pred_config.labels,
                output_dir=FLAGS.output_dir,
                threshold=FLAGS.threshold)
Q
qingqing01 已提交
664 665 666


def predict_video(detector, camera_id):
667
    video_out_name = 'output.mp4'
Q
qingqing01 已提交
668 669 670 671
    if camera_id != -1:
        capture = cv2.VideoCapture(camera_id)
    else:
        capture = cv2.VideoCapture(FLAGS.video_file)
672 673
        video_out_name = os.path.split(FLAGS.video_file)[-1]
    # Get Video info : resolution, fps, frame count
Q
qingqing01 已提交
674 675
    width = int(capture.get(cv2.CAP_PROP_FRAME_WIDTH))
    height = int(capture.get(cv2.CAP_PROP_FRAME_HEIGHT))
676 677 678 679
    fps = int(capture.get(cv2.CAP_PROP_FPS))
    frame_count = int(capture.get(cv2.CAP_PROP_FRAME_COUNT))
    print("fps: %d, frame_count: %d" % (fps, frame_count))

Q
qingqing01 已提交
680 681
    if not os.path.exists(FLAGS.output_dir):
        os.makedirs(FLAGS.output_dir)
682
    out_path = os.path.join(FLAGS.output_dir, video_out_name)
683
    fourcc = cv2.VideoWriter_fourcc(* 'mp4v')
Q
qingqing01 已提交
684 685 686 687 688 689
    writer = cv2.VideoWriter(out_path, fourcc, fps, (width, height))
    index = 1
    while (1):
        ret, frame = capture.read()
        if not ret:
            break
690
        print('detect frame: %d' % (index))
Q
qingqing01 已提交
691
        index += 1
C
cnn 已提交
692
        results = detector.predict([frame], FLAGS.threshold)
Q
qingqing01 已提交
693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708
        im = visualize_box_mask(
            frame,
            results,
            detector.pred_config.labels,
            threshold=FLAGS.threshold)
        im = np.array(im)
        writer.write(im)
        if camera_id != -1:
            cv2.imshow('Mask Detection', im)
            if cv2.waitKey(1) & 0xFF == ord('q'):
                break
    writer.release()


def main():
    pred_config = PredictConfig(FLAGS.model_dir)
709
    detector_func = 'Detector'
G
Guanghua Yu 已提交
710
    if pred_config.arch == 'SOLOv2':
711 712 713 714 715 716 717 718 719 720 721 722 723 724 725
        detector_func = 'DetectorSOLOv2'
    elif pred_config.arch == 'PicoDet':
        detector_func = 'DetectorPicoDet'

    detector = eval(detector_func)(pred_config,
                                   FLAGS.model_dir,
                                   device=FLAGS.device,
                                   run_mode=FLAGS.run_mode,
                                   batch_size=FLAGS.batch_size,
                                   trt_min_shape=FLAGS.trt_min_shape,
                                   trt_max_shape=FLAGS.trt_max_shape,
                                   trt_opt_shape=FLAGS.trt_opt_shape,
                                   trt_calib_mode=FLAGS.trt_calib_mode,
                                   cpu_threads=FLAGS.cpu_threads,
                                   enable_mkldnn=FLAGS.enable_mkldnn)
G
Guanghua Yu 已提交
726

Q
qingqing01 已提交
727
    # predict from video file or camera video stream
G
Guanghua Yu 已提交
728
    if FLAGS.video_file is not None or FLAGS.camera_id != -1:
Q
qingqing01 已提交
729
        predict_video(detector, FLAGS.camera_id)
G
Guanghua Yu 已提交
730 731
    else:
        # predict from image
C
cnn 已提交
732 733
        if FLAGS.image_dir is None and FLAGS.image_file is not None:
            assert FLAGS.batch_size == 1, "batch_size should be 1, when image_file is not None"
G
Guanghua Yu 已提交
734
        img_list = get_test_images(FLAGS.image_dir, FLAGS.image_file)
C
cnn 已提交
735
        predict_image(detector, img_list, FLAGS.batch_size)
G
Guanghua Yu 已提交
736 737 738 739
        if not FLAGS.run_benchmark:
            detector.det_times.info(average=True)
        else:
            mems = {
740 741
                'cpu_rss_mb': detector.cpu_mem / len(img_list),
                'gpu_rss_mb': detector.gpu_mem / len(img_list),
G
Guanghua Yu 已提交
742 743
                'gpu_util': detector.gpu_util * 100 / len(img_list)
            }
744 745 746 747 748

            perf_info = detector.det_times.report(average=True)
            model_dir = FLAGS.model_dir
            mode = FLAGS.run_mode
            model_info = {
749 750
                'model_name': model_dir.strip('/').split('/')[-1],
                'precision': mode.split('_')[-1]
751 752
            }
            data_info = {
753
                'batch_size': FLAGS.batch_size,
754 755 756
                'shape': "dynamic_shape",
                'data_num': perf_info['img_num']
            }
757 758
            det_log = PaddleInferBenchmark(detector.config, model_info,
                                           data_info, perf_info, mems)
759
            det_log('Det')
Q
qingqing01 已提交
760 761 762 763


if __name__ == '__main__':
    paddle.enable_static()
G
Guanghua Yu 已提交
764
    parser = argsparser()
Q
qingqing01 已提交
765 766
    FLAGS = parser.parse_args()
    print_arguments(FLAGS)
G
Guanghua Yu 已提交
767 768 769 770
    FLAGS.device = FLAGS.device.upper()
    assert FLAGS.device in ['CPU', 'GPU', 'XPU'
                            ], "device should be CPU, GPU or XPU"
    assert not FLAGS.use_gpu, "use_gpu has been deprecated, please use --device"
Q
qingqing01 已提交
771 772

    main()