物体检测框架之mmdetection源码阅读（1）

通过readme不难发现单卡训练是以tools/train.py作为程序的入口。首先贴出train.py的源代码。

from __future__ import division
import argparse
import copy
import os
import os.path as osp
import time

import mmcv
import torch
from mmcv import Config
from mmcv.runner import init_dist

from mmdet import __version__
from mmdet.apis import set_random_seed, train_detector
from mmdet.datasets import build_dataset
from mmdet.models import build_detector
from mmdet.utils import collect_env, get_root_logger


def parse_args():
    parser = argparse.ArgumentParser(description='Train a detector')
    parser.add_argument('config', help='train config file path')
    parser.add_argument('--work_dir', help='the dir to save logs and models')
    parser.add_argument(
        '--resume_from', help='the checkpoint file to resume from')
    parser.add_argument(
        '--validate',
        action='store_true',
        help='whether to evaluate the checkpoint during training')
    parser.add_argument(
        '--gpus',
        type=int,
        default=1,
        help='number of gpus to use '
        '(only applicable to non-distributed training)')
    parser.add_argument('--seed', type=int, default=None, help='random seed')
    parser.add_argument(
        '--deterministic',
        action='store_true',
        help='whether to set deterministic options for CUDNN backend.')
    parser.add_argument(
        '--launcher',
        choices=['none', 'pytorch', 'slurm', 'mpi'],
        default='none',
        help='job launcher')
    parser.add_argument('--local_rank', type=int, default=0)
    parser.add_argument(
        '--autoscale-lr',
        action='store_true',
        help='automatically scale lr with the number of gpus')
    args = parser.parse_args()
    if 'LOCAL_RANK' not in os.environ:
        os.environ['LOCAL_RANK'] = str(args.local_rank)

    return args


def main():
    args = parse_args()

    cfg = Config.fromfile(args.config)
    # set cudnn_benchmark
    if cfg.get('cudnn_benchmark', False):
        torch.backends.cudnn.benchmark = True
    # update configs according to CLI args
    if args.work_dir is not None:
        cfg.work_dir = args.work_dir
    if args.resume_from is not None:
        cfg.resume_from = args.resume_from
    cfg.gpus = args.gpus

    if args.autoscale_lr:
        # apply the linear scaling rule (https://arxiv.org/abs/1706.02677)
        cfg.optimizer['lr'] = cfg.optimizer['lr'] * cfg.gpus / 8

    # init distributed env first, since logger depends on the dist info.
    if args.launcher == 'none':
        distributed = False
    else:
        distributed = True
        init_dist(args.launcher, **cfg.dist_params)

    # create work_dir
    mmcv.mkdir_or_exist(osp.abspath(cfg.work_dir))
    # init the logger before other steps
    timestamp = time.strftime('%Y%m%d_%H%M%S', time.localtime())
    log_file = osp.join(cfg.work_dir, '{}.log'.format(timestamp))
    logger = get_root_logger(log_file=log_file, log_level=cfg.log_level)

    # init the meta dict to record some important information such as
    # environment info and seed, which will be logged
    meta = dict()
    # log env info
    env_info_dict = collect_env()
    env_info = '\n'.join([('{}: {}'.format(k, v))
                          for k, v in env_info_dict.items()])
    dash_line = '-' * 60 + '\n'
    logger.info('Environment info:\n' + dash_line + env_info + '\n' +
                dash_line)
    meta['env_info'] = env_info

    # log some basic info
    logger.info('Distributed training: {}'.format(distributed))
    logger.info('Config:\n{}'.format(cfg.text))

    # set random seeds
    if args.seed is not None:
        logger.info('Set random seed to {}, deterministic: {}'.format(
            args.seed, args.deterministic))
        set_random_seed(args.seed, deterministic=args.deterministic)
    cfg.seed = args.seed
    meta['seed'] = args.seed

#通过build_detector()构造模型；
    model = build_detector(
        cfg.model, train_cfg=cfg.train_cfg, test_cfg=cfg.test_cfg)

#通过build_dataset()构造数据集；
    datasets = [build_dataset(cfg.data.train)]
    if len(cfg.workflow) == 2:
        val_dataset = copy.deepcopy(cfg.data.val)
        val_dataset.pipeline = cfg.data.train.pipeline
        datasets.append(build_dataset(val_dataset))
    if cfg.checkpoint_config is not None:
        # save mmdet version, config file content and class names in
        # checkpoints as meta data
        cfg.checkpoint_config.meta = dict(
            mmdet_version=__version__,
            config=cfg.text,
            CLASSES=datasets[0].CLASSES)
    # add an attribute for visualization convenience
    model.CLASSES = datasets[0].CLASSES
# 通过train_detector()训练检测器；
    train_detector(
        model,
        datasets,
        cfg,
        distributed=distributed,
        validate=args.validate,
        timestamp=timestamp,
        meta=meta)


if __name__ == '__main__':
    main()

这段代码主要是通过一个configure文件来设置使用哪个模型，超参的设置；configure文件通过mmcv\mmcv\utils\config.py中的如下代码：

    def fromfile(filename,
                 use_predefined_variables=True,
                 import_custom_modules=True):
        cfg_dict, cfg_text = Config._file2dict(filename,
                                               use_predefined_variables)
        if import_custom_modules and cfg_dict.get('custom_imports', None):
            import_modules_from_strings(**cfg_dict['custom_imports'])
        return Config(cfg_dict, cfg_text=cfg_text, filename=filename)

来解析configure，将其分为字典和字符两块内容。通过conf.model来build_detector()，而conf.model又是什么呢？以下以yolo为例来说明：

_base_ = '../_base_/default_runtime.py'
# model settings
model = dict(
    type='YOLOV3',
    pretrained='open-mmlab://darknet53',
    backbone=dict(type='Darknet', depth=53, out_indices=(3, 4, 5)),
    neck=dict(
        type='YOLOV3Neck',
        num_scales=3,
        in_channels=[1024, 512, 256],
        out_channels=[512, 256, 128]),
    bbox_head=dict(
        type='YOLOV3Head',
        num_classes=80,
        in_channels=[512, 256, 128],
        out_channels=[1024, 512, 256],
        anchor_generator=dict(
            type='YOLOAnchorGenerator',
            base_sizes=[[(116, 90), (156, 198), (373, 326)],
                        [(30, 61), (62, 45), (59, 119)],
                        [(10, 13), (16, 30), (33, 23)]],
            strides=[32, 16, 8]),
        bbox_coder=dict(type='YOLOBBoxCoder'),
        featmap_strides=[32, 16, 8],
        loss_cls=dict(
            type='CrossEntropyLoss',
            use_sigmoid=True,
            loss_weight=1.0,
            reduction='sum'),
        loss_conf=dict(
            type='CrossEntropyLoss',
            use_sigmoid=True,
            loss_weight=1.0,
            reduction='sum'),
        loss_xy=dict(
            type='CrossEntropyLoss',
            use_sigmoid=True,
            loss_weight=2.0,
            reduction='sum'),
        loss_wh=dict(type='MSELoss', loss_weight=2.0, reduction='sum')),
    # training and testing settings
    train_cfg=dict(
        assigner=dict(
            type='GridAssigner',
            pos_iou_thr=0.5,
            neg_iou_thr=0.5,
            min_pos_iou=0)),
    test_cfg=dict(
        nms_pre=1000,
        min_bbox_size=0,
        score_thr=0.05,
        conf_thr=0.005,
        nms=dict(type='nms', iou_threshold=0.45),
        max_per_img=100))
# dataset settings
dataset_type = 'CocoDataset'
data_root = 'data/coco/'
img_norm_cfg = dict(mean=[0, 0, 0], std=[255., 255., 255.], to_rgb=True)
train_pipeline = [
    dict(type='LoadImageFromFile', to_float32=True),
    dict(type='LoadAnnotations', with_bbox=True),
    dict(type='PhotoMetricDistortion'),
    dict(
        type='Expand',
        mean=img_norm_cfg['mean'],
        to_rgb=img_norm_cfg['to_rgb'],
        ratio_range=(1, 2)),
    dict(
        type='MinIoURandomCrop',
        min_ious=(0.4, 0.5, 0.6, 0.7, 0.8, 0.9),
        min_crop_size=0.3),
    dict(type='Resize', img_scale=[(320, 320), (608, 608)], keep_ratio=True),
    dict(type='RandomFlip', flip_ratio=0.5),
    dict(type='Normalize', **img_norm_cfg),
    dict(type='Pad', size_divisor=32),
    dict(type='DefaultFormatBundle'),
    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels'])
]
test_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(
        type='MultiScaleFlipAug',
        img_scale=(608, 608),
        flip=False,
        transforms=[
            dict(type='Resize', keep_ratio=True),
            dict(type='RandomFlip'),
            dict(type='Normalize', **img_norm_cfg),
            dict(type='Pad', size_divisor=32),
            dict(type='ImageToTensor', keys=['img']),
            dict(type='Collect', keys=['img'])
        ])
]
data = dict(
    samples_per_gpu=8,
    workers_per_gpu=4,
    train=dict(
        type=dataset_type,
        ann_file=data_root + 'annotations/instances_train2017.json',
        img_prefix=data_root + 'train2017/',
        pipeline=train_pipeline),
    val=dict(
        type=dataset_type,
        ann_file=data_root + 'annotations/instances_val2017.json',
        img_prefix=data_root + 'val2017/',
        pipeline=test_pipeline),
    test=dict(
        type=dataset_type,
        ann_file=data_root + 'annotations/instances_val2017.json',
        img_prefix=data_root + 'val2017/',
        pipeline=test_pipeline))
# optimizer
optimizer = dict(type='SGD', lr=0.001, momentum=0.9, weight_decay=0.0005)
optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
# learning policy
lr_config = dict(
    policy='step',
    warmup='linear',
    warmup_iters=2000,  # same as burn-in in darknet
    warmup_ratio=0.1,
    step=[218, 246])
# runtime settings
runner = dict(type='EpochBasedRunner', max_epochs=273)
evaluation = dict(interval=1, metric=['bbox'])

conf.model即为上述代码的model字典，通过这个字典来构建detector

然后再通过build_detector()、build_dataset()以及train_detector()来实现构造检测器、构造数据集、训练检测器的功能；

后续分三篇博客依次来讲解这三个函数；