det_infer.py 22.3 KB
Newer Older
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
# Copyright (c) 2021 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
import glob
from functools import reduce

import cv2
import numpy as np
import math

import paddle
from paddle.inference import Config
from paddle.inference import create_predictor

28 29 30 31 32
import sys
# add deploy path of PadleDetection to sys.path
parent_path = os.path.abspath(os.path.join(__file__, *(['..'])))
sys.path.insert(0, parent_path)

33 34
from benchmark_utils import PaddleInferBenchmark
from picodet_postprocess import PicoDetPostProcess
F
Feng Ni 已提交
35
from preprocess import preprocess, Resize, NormalizeImage, Permute, PadStride, LetterBoxResize, Pad, decode_image
W
wangguanzhong 已提交
36 37
from mot.visualize import visualize_box_mask
from mot_utils import argsparser, Timer, get_current_memory_mb
38 39 40 41 42 43 44 45

# Global dictionary
SUPPORT_MODELS = {
    'YOLO',
    'PicoDet',
    'JDE',
    'FairMOT',
    'DeepSORT',
46
    'StrongBaseline',
47 48 49
}


50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66
def bench_log(detector, img_list, model_info, batch_size=1, name=None):
    mems = {
        'cpu_rss_mb': detector.cpu_mem / len(img_list),
        'gpu_rss_mb': detector.gpu_mem / len(img_list),
        'gpu_util': detector.gpu_util * 100 / len(img_list)
    }
    perf_info = detector.det_times.report(average=True)
    data_info = {
        'batch_size': batch_size,
        'shape': "dynamic_shape",
        'data_num': perf_info['img_num']
    }
    log = PaddleInferBenchmark(detector.config, model_info, data_info,
                               perf_info, mems)
    log(name)


67 68 69 70 71 72
class Detector(object):
    """
    Args:
        pred_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
73
        run_mode (str): mode of running(paddle/trt_fp32/trt_fp16)
74 75 76 77 78 79 80 81
        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
82 83
        output_dir (str): The path of output
        threshold (float): The threshold of score for visualization
84 85
    """

86 87 88 89 90 91 92 93 94 95 96 97 98 99 100
    def __init__(
            self,
            model_dir,
            device='CPU',
            run_mode='paddle',
            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,
            output_dir='output',
            threshold=0.5, ):
        self.pred_config = self.set_config(model_dir)
101 102 103 104 105 106 107 108 109 110 111 112 113 114 115
        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
116 117 118 119 120 121
        self.batch_size = batch_size
        self.output_dir = output_dir
        self.threshold = threshold

    def set_config(self, model_dir):
        return PredictConfig(model_dir)
122 123 124 125 126 127 128 129 130 131 132 133 134 135 136

    def preprocess(self, image_list):
        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))

        input_im_lst = []
        input_im_info_lst = []
        for im_path in image_list:
            im, im_info = preprocess(im_path, preprocess_ops)
            input_im_lst.append(im)
            input_im_info_lst.append(im_info)
        inputs = create_inputs(input_im_lst, input_im_info_lst)
137 138 139 140 141
        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]])

142 143
        return inputs

144
    def postprocess(self, inputs, result):
145
        # postprocess output of predictor
146 147 148 149 150 151
        np_boxes_num = result['boxes_num']
        if np_boxes_num[0] <= 0:
            print('[WARNNING] No object detected.')
            result = {'boxes': np.zeros([0, 6]), 'boxes_num': [0]}
        result = {k: v for k, v in result.items() if v is not None}
        return result
152

153
    def predict(self, repeats=1):
154 155
        '''
        Args:
156
            repeats (int): repeats number for prediction
157
        Returns:
158
            result (dict): include 'boxes': np.ndarray: shape:[N,6], N: number of box,
159 160
                            matix element:[class, score, x_min, y_min, x_max, y_max]
        '''
161 162
        # model prediction
        np_boxes, np_boxes_num = None, None
163 164 165 166 167 168 169
        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()
            boxes_num = self.predictor.get_output_handle(output_names[1])
            np_boxes_num = boxes_num.copy_to_cpu()
170 171 172 173 174 175 176 177 178 179 180 181 182
        result = dict(boxes=np_boxes, boxes_num=np_boxes_num)
        return result

    def merge_batch_result(self, batch_result):
        if len(batch_result) == 1:
            return batch_result[0]
        res_key = batch_result[0].keys()
        results = {k: [] for k in res_key}
        for res in batch_result:
            for k, v in res.items():
                results[k].append(v)
        for k, v in results.items():
            results[k] = np.concatenate(v)
183 184 185 186 187
        return results

    def get_timer(self):
        return self.det_times

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
    def predict_image(self,
                      image_list,
                      run_benchmark=False,
                      repeats=1,
                      visual=True):
        batch_loop_cnt = math.ceil(float(len(image_list)) / self.batch_size)
        results = []
        for i in range(batch_loop_cnt):
            start_index = i * self.batch_size
            end_index = min((i + 1) * self.batch_size, len(image_list))
            batch_image_list = image_list[start_index:end_index]
            if run_benchmark:
                # preprocess
                inputs = self.preprocess(batch_image_list)  # warmup
                self.det_times.preprocess_time_s.start()
                inputs = self.preprocess(batch_image_list)
                self.det_times.preprocess_time_s.end()

                # model prediction
                result = self.predict(repeats=repeats)  # warmup
                self.det_times.inference_time_s.start()
                result = self.predict(repeats=repeats)
                self.det_times.inference_time_s.end(repeats=repeats)

                # postprocess
                result_warmup = self.postprocess(inputs, result)  # warmup
                self.det_times.postprocess_time_s.start()
                result = self.postprocess(inputs, result)
                self.det_times.postprocess_time_s.end()
                self.det_times.img_num += len(batch_image_list)

                cm, gm, gu = get_current_memory_mb()
                self.cpu_mem += cm
                self.gpu_mem += gm
                self.gpu_util += gu
            else:
                # preprocess
                self.det_times.preprocess_time_s.start()
                inputs = self.preprocess(batch_image_list)
                self.det_times.preprocess_time_s.end()

                # model prediction
                self.det_times.inference_time_s.start()
                result = self.predict()
                self.det_times.inference_time_s.end()

                # postprocess
                self.det_times.postprocess_time_s.start()
                result = self.postprocess(inputs, result)
                self.det_times.postprocess_time_s.end()
                self.det_times.img_num += len(batch_image_list)

                if visual:
                    visualize(
                        batch_image_list,
                        result,
                        self.pred_config.labels,
                        output_dir=self.output_dir,
                        threshold=self.threshold)

            results.append(result)
            if visual:
                print('Test iter {}'.format(i))

        results = self.merge_batch_result(results)
        return results
254

255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271
    def predict_video(self, video_file, camera_id):
        video_out_name = 'output.mp4'
        if camera_id != -1:
            capture = cv2.VideoCapture(camera_id)
        else:
            capture = cv2.VideoCapture(video_file)
            video_out_name = os.path.split(video_file)[-1]
        # Get Video info : resolution, fps, frame count
        width = int(capture.get(cv2.CAP_PROP_FRAME_WIDTH))
        height = int(capture.get(cv2.CAP_PROP_FRAME_HEIGHT))
        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))

        if not os.path.exists(self.output_dir):
            os.makedirs(self.output_dir)
        out_path = os.path.join(self.output_dir, video_out_name)
W
wangguanzhong 已提交
272
        fourcc = cv2.VideoWriter_fourcc(* 'mp4v')
273 274 275 276 277 278 279 280 281
        writer = cv2.VideoWriter(out_path, fourcc, fps, (width, height))
        index = 1
        while (1):
            ret, frame = capture.read()
            if not ret:
                break
            print('detect frame: %d' % (index))
            index += 1
            results = self.predict_image([frame], visual=False)
282

283 284 285 286 287 288 289 290 291 292 293 294
            im = visualize_box_mask(
                frame,
                results,
                self.pred_config.labels,
                threshold=self.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()
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


def create_inputs(imgs, im_info):
    """generate input for different model type
    Args:
        imgs (list(numpy)): list of images (np.ndarray)
        im_info (list(dict)): list of image info
    Returns:
        inputs (dict): input of model
    """
    inputs = {}

    im_shape = []
    scale_factor = []
    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

    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'))

    inputs['im_shape'] = np.concatenate(im_shape, axis=0)
    inputs['scale_factor'] = np.concatenate(scale_factor, axis=0)

    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)
    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']
        self.mask = False
        self.use_dynamic_shape = yml_conf['use_dynamic_shape']
        if 'mask' in yml_conf:
            self.mask = yml_conf['mask']
        self.tracker = None
        if 'tracker' in yml_conf:
            self.tracker = yml_conf['tracker']
        if 'NMS' in yml_conf:
            self.nms = yml_conf['NMS']
        if 'fpn_stride' in yml_conf:
            self.fpn_stride = yml_conf['fpn_stride']
        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,
388
                   run_mode='paddle',
389 390 391 392 393 394 395 396 397 398 399 400 401 402
                   batch_size=1,
                   device='CPU',
                   min_subgraph_size=3,
                   use_dynamic_shape=False,
                   trt_min_shape=1,
                   trt_max_shape=1280,
                   trt_opt_shape=640,
                   trt_calib_mode=False,
                   cpu_threads=1,
                   enable_mkldnn=False):
    """set AnalysisConfig, generate AnalysisPredictor
    Args:
        model_dir (str): root path of __model__ and __params__
        device (str): Choose the device you want to run, it can be: CPU/GPU/XPU, default is CPU
403
        run_mode (str): mode of running(paddle/trt_fp32/trt_fp16/trt_int8)
404 405 406 407 408 409 410 411 412 413 414
        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
        trt_calib_mode (bool): If the model is produced by TRT offline quantitative
            calibration, trt_calib_mode need to set True
    Returns:
        predictor (PaddlePredictor): AnalysisPredictor
    Raises:
        ValueError: predict by TensorRT need device == 'GPU'.
    """
415
    if device != 'GPU' and run_mode != 'paddle':
416 417 418
        raise ValueError(
            "Predict by TensorRT mode: {}, expect device=='GPU', but device == {}"
            .format(run_mode, device))
419 420 421 422 423 424 425 426 427
    infer_model = os.path.join(model_dir, 'model.pdmodel')
    infer_params = os.path.join(model_dir, 'model.pdiparams')
    if not os.path.exists(infer_model):
        infer_model = os.path.join(model_dir, 'inference.pdmodel')
        infer_params = os.path.join(model_dir, 'inference.pdiparams')
        if not os.path.exists(infer_model):
            raise ValueError(
                "Cannot find any inference model in dir: {},".format(model_dir))
    config = Config(infer_model, infer_params)
428 429 430 431 432 433
    if device == 'GPU':
        # initial GPU memory(M), device ID
        config.enable_use_gpu(200, 0)
        # optimize graph and fuse op
        config.switch_ir_optim(True)
    elif device == 'XPU':
434
        config.enable_lite_engine()
435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456
        config.enable_xpu(10 * 1024 * 1024)
    else:
        config.disable_gpu()
        config.set_cpu_math_library_num_threads(cpu_threads)
        if enable_mkldnn:
            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

    precision_map = {
        'trt_int8': Config.Precision.Int8,
        'trt_fp32': Config.Precision.Float32,
        'trt_fp16': Config.Precision.Half
    }
    if run_mode in precision_map.keys():
        config.enable_tensorrt_engine(
457
            workspace_size=1 << 25,
458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518
            max_batch_size=batch_size,
            min_subgraph_size=min_subgraph_size,
            precision_mode=precision_map[run_mode],
            use_static=False,
            use_calib_mode=trt_calib_mode)

        if use_dynamic_shape:
            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]
            }
            config.set_trt_dynamic_shape_info(min_input_shape, max_input_shape,
                                              opt_input_shape)
            print('trt set dynamic shape done!')

    # 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)
    return predictor, config


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


519
def visualize(image_list, result, labels, output_dir='output/', threshold=0.5):
520 521 522
    # visualize the predict result
    start_idx = 0
    for idx, image_file in enumerate(image_list):
523
        im_bboxes_num = result['boxes_num'][idx]
524
        im_results = {}
525 526 527
        if 'boxes' in result:
            im_results['boxes'] = result['boxes'][start_idx:start_idx +
                                                  im_bboxes_num, :]
528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546
        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)


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


def main():
547 548 549 550
    deploy_file = os.path.join(FLAGS.model_dir, 'infer_cfg.yml')
    with open(deploy_file) as f:
        yml_conf = yaml.safe_load(f)
    arch = yml_conf['arch']
551
    detector_func = 'Detector'
552
    detector = eval(detector_func)(FLAGS.model_dir,
553 554 555 556 557 558 559 560
                                   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,
561 562 563
                                   enable_mkldnn=FLAGS.enable_mkldnn,
                                   threshold=FLAGS.threshold,
                                   output_dir=FLAGS.output_dir)
564 565 566

    # predict from video file or camera video stream
    if FLAGS.video_file is not None or FLAGS.camera_id != -1:
567
        detector.predict_video(FLAGS.video_file, FLAGS.camera_id)
568 569 570 571 572
    else:
        # predict from image
        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"
        img_list = get_test_images(FLAGS.image_dir, FLAGS.image_file)
573
        detector.predict_image(img_list, FLAGS.run_benchmark, repeats=10)
574 575 576 577
        if not FLAGS.run_benchmark:
            detector.det_times.info(average=True)
        else:
            mode = FLAGS.run_mode
578
            model_dir = FLAGS.model_dir
579 580 581 582
            model_info = {
                'model_name': model_dir.strip('/').split('/')[-1],
                'precision': mode.split('_')[-1]
            }
583
            bench_log(detector, img_list, model_info, name='DET')
584 585 586 587 588 589 590 591 592 593 594


if __name__ == '__main__':
    paddle.enable_static()
    parser = argsparser()
    FLAGS = parser.parse_args()
    print_arguments(FLAGS)
    FLAGS.device = FLAGS.device.upper()
    assert FLAGS.device in ['CPU', 'GPU', 'XPU'
                            ], "device should be CPU, GPU or XPU"
    main()