Target Detection-1-目标检测最新网址、论文、资料

Yolov-1-TX2上用YOLOv3训练自己数据集的流程(VOC2007-TX2-GPU)

Yolov--2--一文全面了解深度学习性能优化加速引擎---TensorRT

Yolov--3--TensorRT中yolov3性能优化加速(基于caffe)

yolov-5-目标检测:YOLOv2算法原理详解

yolov--8--Tensorflow实现YOLO v3

yolov--9--YOLO v3的剪枝优化

yolov--10--目标检测模型的参数评估指标详解、概念解析

yolov--11--YOLO v3的原版训练记录、mAP、AP、recall、precision、time等评价指标计算

yolov--12--YOLOv3的原理深度剖析和关键点讲解


1、arXiv: https://arxiv.org/abs/1902.04103

github: https://github.com/dmlc/gluon-cv

作者团队:Amazon Web Services

注:2019年02月11日刚出炉的paper

目标检测训练方法:

https://github.com/dmlc/gluon-cv.git

https://mp.weixin.qq.com/s/u1s3vh4Gnk5WyArQVwaFBw

https://mp.weixin.qq.com/s/6VfxXnm_BERubvCTWLfHBg


剪枝优化:

https://github.com/talebolano/yolov3-network-slimming

 

https://download.csdn.net/download/qq_33350808/10731748


懒人学AI:

yolov1:   https://mp.weixin.qq.com/s/cwAhC6z5MO_QPm-VeFLF2w

yolov2: https://mp.weixin.qq.com/s/eYG9hd-HgjiyEikozM5_Xw

yolov3: https://mp.weixin.qq.com/s/t9V0610HXedFSDTcHsckIw


基于Objects as Points检测算法的小小尝试

https://mp.weixin.qq.com/s/j9KU11Lq1PHh0tqHsDvi5A

 


 

https://mp.weixin.qq.com/s/JZbMYjEjRypvArYZrOIWpg


 2019-7-9

object-detection

Mask_RCNN

https://github.com/topics/object-detection?o=desc&s=stars

deep learning object detection

Target Detection-1-目标检测最新网址、论文、资料_第1张图片

https://github.com/hoya012/deep_learning_object_detection


2、YOLOv3解释

下面是一些可供利用的参考资料:

https://xmfbit.github.io/2018/04/01/paper-yolov3/

  • YOLO的项目主页Darknet YOLO
  • 作者主页上的paper链接
  • 知乎专栏上的全文翻译
  • FPN论文Feature pyramid networks for object detection
  • 知乎上的解答:AP是什么,怎么计算

https://www.cnblogs.com/makefile/p/YOLOv3.html

http://www.photoplayhouse.com/?s=yolov3%20%E9%BB%98%E8%AE%A4%E6%A1%86

 

3、YOLO-v3改进

https://gluon-cv.mxnet.io/model_zoo/detection.html

The training commands work with the following scripts:

  • For SSD [1] networks: Download train_ssd.py
  • For Faster-RCNN [2] networks: Download train_faster_rcnn.py
  • For YOLO v3 [3] networks: Download train_yolo3.py
  • Pascal VOC

    HINT

    For Pascal VOC dataset, training image set is the union of 2007trainval and 2012trainval and validation image set is 2007test. 

    The VOC metric, mean Average Precision (mAP) across all classes with IoU threshold 0.5 is reported.

YOLO-v3 

YOLO-v3 models can be evaluated and used for prediction at different resolutions. Different mAPs are reported with various evaluation resolutions, however, the models are identical.

Checkout YOLO demo tutorial here: 03. Predict with pre-trained YOLO models

yolo3_darknet53_voc [3] (320x320) 79.3 shell script log
yolo3_darknet53_voc [3] (416x416) 81.5 shell script log
yolo3_mobilenet1.0_voc [3] (320x320) 73.9 shell script log
yolo3_mobilenet1.0_voc [3] (416x416) 75.8 shell script log

1、 

python3 train_yolo3.py --network darknet53 --dataset voc --gpus 0,1,2,3,4,5,6,7 --batch-size 64 -j 16 --log-interval 100 --lr-decay-epoch 160,180 --epochs 200 --syncbn --warmup-epochs 4

2、

python3 train_yolo3.py --network darknet53 --dataset voc --gpus 0,1,2,3,4,5,6,7 --batch-size 64 -j 16 --log-interval 100 --lr-decay-epoch 160,180 --epochs 200 --syncbn --warmup-epochs 4

3、

python3 train_yolo3.py --network mobilenet1.0 --dataset voc --gpus 0,1,2,3,4,5,6,7 --batch-size 64 -j 32 --log-interval 100 --lr-decay-epoch 160,180 --epochs 200 --syncbn --warmup-epochs 4

4、

python3 train_yolo3.py --network mobilenet1.0 --dataset voc --gpus 0,1,2,3,4,5,6,7 --batch-size 64 -j 32 --log-interval 100 --lr-decay-epoch 160,180 --epochs 200 --syncbn --warmup-epochs 4

1、https://gluon-cv.mxnet.io/build/examples_detection/train_yolo_v3.html

 

For YOLO v3 [3] networks: Download train_yolo3.py

"""Train YOLOv3 with random shapes."""
import argparse
import os
import logging
import time
import warnings
import numpy as np
import mxnet as mx
from mxnet import nd
from mxnet import gluon
from mxnet import autograd
import gluoncv as gcv
from gluoncv import data as gdata
from gluoncv import utils as gutils
from gluoncv.model_zoo import get_model
from gluoncv.data.batchify import Tuple, Stack, Pad
from gluoncv.data.transforms.presets.yolo import YOLO3DefaultTrainTransform
from gluoncv.data.transforms.presets.yolo import YOLO3DefaultValTransform
from gluoncv.data.dataloader import RandomTransformDataLoader
from gluoncv.utils.metrics.voc_detection import VOC07MApMetric
from gluoncv.utils.metrics.coco_detection import COCODetectionMetric
from gluoncv.utils import LRScheduler

def parse_args():
    parser = argparse.ArgumentParser(description='Train YOLO networks with random input shape.')
    parser.add_argument('--network', type=str, default='darknet53',
                        help="Base network name which serves as feature extraction base.")
    parser.add_argument('--data-shape', type=int, default=416,
                        help="Input data shape for evaluation, use 320, 416, 608... " +
                             "Training is with random shapes from (320 to 608).")
    parser.add_argument('--batch-size', type=int, default=64,
                        help='Training mini-batch size')
    parser.add_argument('--dataset', type=str, default='voc',
                        help='Training dataset. Now support voc.')
    parser.add_argument('--num-workers', '-j', dest='num_workers', type=int,
                        default=4, help='Number of data workers, you can use larger '
                        'number to accelerate data loading, if you CPU and GPUs are powerful.')
    parser.add_argument('--gpus', type=str, default='0',
                        help='Training with GPUs, you can specify 1,3 for example.')
    parser.add_argument('--epochs', type=int, default=200,
                        help='Training epochs.')
    parser.add_argument('--resume', type=str, default='',
                        help='Resume from previously saved parameters if not None. '
                        'For example, you can resume from ./yolo3_xxx_0123.params')
    parser.add_argument('--start-epoch', type=int, default=0,
                        help='Starting epoch for resuming, default is 0 for new training.'
                        'You can specify it to 100 for example to start from 100 epoch.')
    parser.add_argument('--lr', type=float, default=0.001,
                        help='Learning rate, default is 0.001')
    parser.add_argument('--lr-mode', type=str, default='step',
                        help='learning rate scheduler mode. options are step, poly and cosine.')
    parser.add_argument('--lr-decay', type=float, default=0.1,
                        help='decay rate of learning rate. default is 0.1.')
    parser.add_argument('--lr-decay-period', type=int, default=0,
                        help='interval for periodic learning rate decays. default is 0 to disable.')
    parser.add_argument('--lr-decay-epoch', type=str, default='160,180',
                        help='epochs at which learning rate decays. default is 160,180.')
    parser.add_argument('--warmup-lr', type=float, default=0.0,
                        help='starting warmup learning rate. default is 0.0.')
    parser.add_argument('--warmup-epochs', type=int, default=0,
                        help='number of warmup epochs.')
    parser.add_argument('--momentum', type=float, default=0.9,
                        help='SGD momentum, default is 0.9')
    parser.add_argument('--wd', type=float, default=0.0005,
                        help='Weight decay, default is 5e-4')
    parser.add_argument('--log-interval', type=int, default=100,
                        help='Logging mini-batch interval. Default is 100.')
    parser.add_argument('--save-prefix', type=str, default='',
                        help='Saving parameter prefix')
    parser.add_argument('--save-interval', type=int, default=10,
                        help='Saving parameters epoch interval, best model will always be saved.')
    parser.add_argument('--val-interval', type=int, default=1,
                        help='Epoch interval for validation, increase the number will reduce the '
                             'training time if validation is slow.')
    parser.add_argument('--seed', type=int, default=233,
                        help='Random seed to be fixed.')
    parser.add_argument('--num-samples', type=int, default=-1,
                        help='Training images. Use -1 to automatically get the number.')
    parser.add_argument('--syncbn', action='store_true',
                        help='Use synchronize BN across devices.')
    parser.add_argument('--no-random-shape', action='store_true',
                        help='Use fixed size(data-shape) throughout the training, which will be faster '
                        'and require less memory. However, final model will be slightly worse.')
    parser.add_argument('--no-wd', action='store_true',
                        help='whether to remove weight decay on bias, and beta/gamma for batchnorm layers.')
    parser.add_argument('--mixup', action='store_true',
                        help='whether to enable mixup.')
    parser.add_argument('--no-mixup-epochs', type=int, default=20,
                        help='Disable mixup training if enabled in the last N epochs.')
    parser.add_argument('--label-smooth', action='store_true', help='Use label smoothing.')
    args = parser.parse_args()
    return args

def get_dataset(dataset, args):
    if dataset.lower() == 'voc':
        train_dataset = gdata.VOCDetection(
            splits=[(2007, 'trainval'), (2012, 'trainval')])
        val_dataset = gdata.VOCDetection(
            splits=[(2007, 'test')])
        val_metric = VOC07MApMetric(iou_thresh=0.5, class_names=val_dataset.classes)
    elif dataset.lower() == 'coco':
        train_dataset = gdata.COCODetection(splits='instances_train2017', use_crowd=False)
        val_dataset = gdata.COCODetection(splits='instances_val2017', skip_empty=False)
        val_metric = COCODetectionMetric(
            val_dataset, args.save_prefix + '_eval', cleanup=True,
            data_shape=(args.data_shape, args.data_shape))
    else:
        raise NotImplementedError('Dataset: {} not implemented.'.format(dataset))
    if args.num_samples < 0:
        args.num_samples = len(train_dataset)
    if args.mixup:
        from gluoncv.data import MixupDetection
        train_dataset = MixupDetection(train_dataset)
    return train_dataset, val_dataset, val_metric

def get_dataloader(net, train_dataset, val_dataset, data_shape, batch_size, num_workers, args):
    """Get dataloader."""
    width, height = data_shape, data_shape
    batchify_fn = Tuple(*([Stack() for _ in range(6)] + [Pad(axis=0, pad_val=-1) for _ in range(1)]))  # stack image, all targets generated
    if args.no_random_shape:
        train_loader = gluon.data.DataLoader(
            train_dataset.transform(YOLO3DefaultTrainTransform(width, height, net, mixup=args.mixup)),
            batch_size, True, batchify_fn=batchify_fn, last_batch='rollover', num_workers=num_workers)
    else:
        transform_fns = [YOLO3DefaultTrainTransform(x * 32, x * 32, net, mixup=args.mixup) for x in range(10, 20)]
        train_loader = RandomTransformDataLoader(
            transform_fns, train_dataset, batch_size=batch_size, interval=10, last_batch='rollover',
            shuffle=True, batchify_fn=batchify_fn, num_workers=num_workers)
    val_batchify_fn = Tuple(Stack(), Pad(pad_val=-1))
    val_loader = gluon.data.DataLoader(
        val_dataset.transform(YOLO3DefaultValTransform(width, height)),
        batch_size, False, batchify_fn=val_batchify_fn, last_batch='keep', num_workers=num_workers)
    return train_loader, val_loader

def save_params(net, best_map, current_map, epoch, save_interval, prefix):
    current_map = float(current_map)
    if current_map > best_map[0]:
        best_map[0] = current_map
        net.save_parameters('{:s}_best.params'.format(prefix, epoch, current_map))
        with open(prefix+'_best_map.log', 'a') as f:
            f.write('{:04d}:\t{:.4f}\n'.format(epoch, current_map))
    if save_interval and epoch % save_interval == 0:
        net.save_parameters('{:s}_{:04d}_{:.4f}.params'.format(prefix, epoch, current_map))

def validate(net, val_data, ctx, eval_metric):
    """Test on validation dataset."""
    eval_metric.reset()
    # set nms threshold and topk constraint
    net.set_nms(nms_thresh=0.45, nms_topk=400)
    mx.nd.waitall()
    net.hybridize()
    for batch in val_data:
        data = gluon.utils.split_and_load(batch[0], ctx_list=ctx, batch_axis=0, even_split=False)
        label = gluon.utils.split_and_load(batch[1], ctx_list=ctx, batch_axis=0, even_split=False)
        det_bboxes = []
        det_ids = []
        det_scores = []
        gt_bboxes = []
        gt_ids = []
        gt_difficults = []
        for x, y in zip(data, label):
            # get prediction results
            ids, scores, bboxes = net(x)
            det_ids.append(ids)
            det_scores.append(scores)
            # clip to image size
            det_bboxes.append(bboxes.clip(0, batch[0].shape[2]))
            # split ground truths
            gt_ids.append(y.slice_axis(axis=-1, begin=4, end=5))
            gt_bboxes.append(y.slice_axis(axis=-1, begin=0, end=4))
            gt_difficults.append(y.slice_axis(axis=-1, begin=5, end=6) if y.shape[-1] > 5 else None)

        # update metric
        eval_metric.update(det_bboxes, det_ids, det_scores, gt_bboxes, gt_ids, gt_difficults)
    return eval_metric.get()

def train(net, train_data, val_data, eval_metric, ctx, args):
    """Training pipeline"""
    net.collect_params().reset_ctx(ctx)
    if args.no_wd:
        for k, v in net.collect_params('.*beta|.*gamma|.*bias').items():
            v.wd_mult = 0.0

    if args.label_smooth:
        net._target_generator._label_smooth = True

    if args.lr_decay_period > 0:
        lr_decay_epoch = list(range(args.lr_decay_period, args.epochs, args.lr_decay_period))
    else:
        lr_decay_epoch = [int(i) for i in args.lr_decay_epoch.split(',')]
    lr_scheduler = LRScheduler(mode=args.lr_mode,
                               baselr=args.lr,
                               niters=args.num_samples // args.batch_size,
                               nepochs=args.epochs,
                               step=lr_decay_epoch,
                               step_factor=args.lr_decay, power=2,
                               warmup_epochs=args.warmup_epochs)

    trainer = gluon.Trainer(
        net.collect_params(), 'sgd',
        {'wd': args.wd, 'momentum': args.momentum, 'lr_scheduler': lr_scheduler},
        kvstore='local')

    # targets
    sigmoid_ce = gluon.loss.SigmoidBinaryCrossEntropyLoss(from_sigmoid=False)
    l1_loss = gluon.loss.L1Loss()

    # metrics
    obj_metrics = mx.metric.Loss('ObjLoss')
    center_metrics = mx.metric.Loss('BoxCenterLoss')
    scale_metrics = mx.metric.Loss('BoxScaleLoss')
    cls_metrics = mx.metric.Loss('ClassLoss')

    # set up logger
    logging.basicConfig()
    logger = logging.getLogger()
    logger.setLevel(logging.INFO)
    log_file_path = args.save_prefix + '_train.log'
    log_dir = os.path.dirname(log_file_path)
    if log_dir and not os.path.exists(log_dir):
        os.makedirs(log_dir)
    fh = logging.FileHandler(log_file_path)
    logger.addHandler(fh)
    logger.info(args)
    logger.info('Start training from [Epoch {}]'.format(args.start_epoch))
    best_map = [0]
    for epoch in range(args.start_epoch, args.epochs):
        if args.mixup:
            # TODO(zhreshold): more elegant way to control mixup during runtime
            try:
                train_data._dataset.set_mixup(np.random.beta, 1.5, 1.5)
            except AttributeError:
                train_data._dataset._data.set_mixup(np.random.beta, 1.5, 1.5)
            if epoch >= args.epochs - args.no_mixup_epochs:
                try:
                    train_data._dataset.set_mixup(None)
                except AttributeError:
                    train_data._dataset._data.set_mixup(None)

        tic = time.time()
        btic = time.time()
        mx.nd.waitall()
        net.hybridize()
        for i, batch in enumerate(train_data):
            batch_size = batch[0].shape[0]
            data = gluon.utils.split_and_load(batch[0], ctx_list=ctx, batch_axis=0)
            # objectness, center_targets, scale_targets, weights, class_targets
            fixed_targets = [gluon.utils.split_and_load(batch[it], ctx_list=ctx, batch_axis=0) for it in range(1, 6)]
            gt_boxes = gluon.utils.split_and_load(batch[6], ctx_list=ctx, batch_axis=0)
            sum_losses = []
            obj_losses = []
            center_losses = []
            scale_losses = []
            cls_losses = []
            with autograd.record():
                for ix, x in enumerate(data):
                    obj_loss, center_loss, scale_loss, cls_loss = net(x, gt_boxes[ix], *[ft[ix] for ft in fixed_targets])
                    sum_losses.append(obj_loss + center_loss + scale_loss + cls_loss)
                    obj_losses.append(obj_loss)
                    center_losses.append(center_loss)
                    scale_losses.append(scale_loss)
                    cls_losses.append(cls_loss)
                autograd.backward(sum_losses)
            lr_scheduler.update(i, epoch)
            trainer.step(batch_size)
            obj_metrics.update(0, obj_losses)
            center_metrics.update(0, center_losses)
            scale_metrics.update(0, scale_losses)
            cls_metrics.update(0, cls_losses)
            if args.log_interval and not (i + 1) % args.log_interval:
                name1, loss1 = obj_metrics.get()
                name2, loss2 = center_metrics.get()
                name3, loss3 = scale_metrics.get()
                name4, loss4 = cls_metrics.get()
                logger.info('[Epoch {}][Batch {}], LR: {:.2E}, Speed: {:.3f} samples/sec, {}={:.3f}, {}={:.3f}, {}={:.3f}, {}={:.3f}'.format(
                    epoch, i, trainer.learning_rate, batch_size/(time.time()-btic), name1, loss1, name2, loss2, name3, loss3, name4, loss4))
            btic = time.time()

        name1, loss1 = obj_metrics.get()
        name2, loss2 = center_metrics.get()
        name3, loss3 = scale_metrics.get()
        name4, loss4 = cls_metrics.get()
        logger.info('[Epoch {}] Training cost: {:.3f}, {}={:.3f}, {}={:.3f}, {}={:.3f}, {}={:.3f}'.format(
            epoch, (time.time()-tic), name1, loss1, name2, loss2, name3, loss3, name4, loss4))
        if not (epoch + 1) % args.val_interval:
            # consider reduce the frequency of validation to save time
            map_name, mean_ap = validate(net, val_data, ctx, eval_metric)
            val_msg = '\n'.join(['{}={}'.format(k, v) for k, v in zip(map_name, mean_ap)])
            logger.info('[Epoch {}] Validation: \n{}'.format(epoch, val_msg))
            current_map = float(mean_ap[-1])
        else:
            current_map = 0.
        save_params(net, best_map, current_map, epoch, args.save_interval, args.save_prefix)

if __name__ == '__main__':
    args = parse_args()
    # fix seed for mxnet, numpy and python builtin random generator.
    gutils.random.seed(args.seed)

    # training contexts
    ctx = [mx.gpu(int(i)) for i in args.gpus.split(',') if i.strip()]
    ctx = ctx if ctx else [mx.cpu()]

    # network
    net_name = '_'.join(('yolo3', args.network, args.dataset))
    args.save_prefix += net_name
    # use sync bn if specified
    if args.syncbn and len(ctx) > 1:
        net = get_model(net_name, pretrained_base=True, norm_layer=gluon.contrib.nn.SyncBatchNorm,
                        norm_kwargs={'num_devices': len(ctx)})
        async_net = get_model(net_name, pretrained_base=False)  # used by cpu worker
    else:
        net = get_model(net_name, pretrained_base=True)
        async_net = net
    if args.resume.strip():
        net.load_parameters(args.resume.strip())
        async_net.load_parameters(args.resume.strip())
    else:
        with warnings.catch_warnings(record=True) as w:
            warnings.simplefilter("always")
            net.initialize()
            async_net.initialize()

    # training data
    train_dataset, val_dataset, eval_metric = get_dataset(args.dataset, args)
    train_data, val_data = get_dataloader(
        async_net, train_dataset, val_dataset, args.data_shape, args.batch_size, args.num_workers, args)

    # training
    train(net, train_data, val_data, eval_metric, ctx, args)

 

你可能感兴趣的:(Object,Detection)