mmdetection(一）安装及训练、测试VOC格式的数据

一、安装

         https://github.com/open-mmlab/mmdetection/blob/master/docs/INSTALL.md

 

二、训练自己的数据

　1、数据　

     mmdet的默认格式是coco的，这里就以voc格式为例，data下文件夹摆放位置如图

 

   2、训练

         （1）修改configs文件下的文件

　　         可先复制一份，然后自己命名一下。比如retinanet_x101_64x4d_fpn_1x.py，修改的部分主要是dataset settings部分，这部分可直接参考

         pascal_voc/faster_rcnn_r50_fpn_1x_voc0712.py（如下）;还有一部分是修改该文件下的num_classes（类别数+1）

# dataset settings
dataset_type = \'VOCDataset\'
data_root = \'data/VOCdevkit/\'
img_norm_cfg = dict(
    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
    dict(type=\'LoadImageFromFile\'),
    dict(type=\'LoadAnnotations\', with_bbox=True),
    dict(type=\'Resize\', img_scale=(1000, 600), 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=(1000, 600),
        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(
    imgs_per_gpu=2,
    workers_per_gpu=2,
    train=dict(
        type=\'RepeatDataset\',
        times=3,
        dataset=dict(
            type=dataset_type,
            ann_file=[
                data_root + \'VOC2007/ImageSets/Main/trainval.txt\'
                
            ],
            img_prefix=[data_root + \'VOC2007/\'],
            pipeline=train_pipeline)),
    val=dict(
        type=dataset_type,
        ann_file=data_root + \'VOC2007/ImageSets/Main/test.txt\',
        img_prefix=data_root + \'VOC2007/\',
        pipeline=test_pipeline),
    test=dict(
        type=dataset_type,
        ann_file=data_root + \'VOC2007/ImageSets/Main/test.txt\',
        img_prefix=data_root + \'VOC2007/\',
        pipeline=test_pipeline))
evaluation = dict(interval=1, metric=\'mAP\')
# optimizer
optimizer = dict(type=\'SGD\', lr=0.01, momentum=0.9, weight_decay=0.0001)
optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
# learning policy
lr_config = dict(policy=\'step\', step=[3])  # actual epoch = 3 * 3 = 9
checkpoint_config = dict(interval=1)
# yapf:disable
log_config = dict(
    interval=50,
    hooks=[
        dict(type=\'TextLoggerHook\'),
        # dict(type=\'TensorboardLoggerHook\')
    ])
# yapf:enable
# runtime settings
total_epochs = 100  # actual epoch = 4 * 3 = 12
dist_params = dict(backend=\'nccl\')
log_level = \'INFO\'
work_dir = \'./work_dirs/faster_rcnn_r50_fpn_1x_voc0712\'
load_from = None
resume_from = None
workflow = [(\'train\', 1)]

View Code

 

          （2）修改mmdet/datasets/voc.py下classes为自己的类

          （3）训练

　　　　　　python tools/train.py configs/pascal_voc/faster_rcnn_r50_fpn_1x_voc0712_my.py 

 3、测试

      （1）输出mAP

 　   修改mmdetection/mmdet/core/evaluation   voc_classes()返回自己的类

        python3 tools/test.py configs/pascal_voc/faster_rcnn_r50_fpn_1x_voc0712_my.py work_dirs/faster_rcnn_r50_fpn_1x_voc0712/latest.pth --eval mAP    --show

 　  (2)  测试单张图片

        参考 demo/webcam_demo.py，

python demo/img_demo.py configs/pascal_voc/faster_rcnn_r50_fpn_1x_voc0712_my.py work_dirs/faster_rcnn_r50_fpn_1x_voc0712/latest.pth demo/2017-09-05-161908.jpg

 

        

import argparse
import torch

from mmdet.apis import inference_detector, init_detector, show_result


def parse_args():
    parser = argparse.ArgumentParser(description=\'MMDetection image demo\')
    parser.add_argument(\'config\', help=\'test config file path\')
    parser.add_argument(\'checkpoint\', help=\'checkpoint file\')
    parser.add_argument(\'imagepath\', help=\'the path of image to test\')
    parser.add_argument(\'--device\', type=int, default=0, help=\'CUDA device id\')
    parser.add_argument(
        \'--score-thr\', type=float, default=0.5, help=\'bbox score threshold\')
    args = parser.parse_args()
    return args


def main():
    args = parse_args()

    model = init_detector(
        args.config, args.checkpoint, device=torch.device(\'cuda\', args.device))

    result = inference_detector(model, args.imagepath)
 
    show_result(
        args.imagepath, result, model.CLASSES, score_thr=args.score_thr, wait_time=0)


if __name__ == \'__main__\':
    main()

 

参考

https://zhuanlan.zhihu.com/p/101202864

https://blog.csdn.net/laizi_laizi/article/details/104256781