OpenMMLab AI实战训练营day6MMDetection代码教程

模型训练

自己写的配置文件参考下

_base_ = 'E:\github\MyOpenMMLab\homework_3_mmdetection\mmdetection\configs\\rtmdet\\rtmdet_tiny_8xb32-300e_coco.py'

data_root = 'E:\github\MyOpenMMLab\homework_3_mmdetection\data\Drink_284_Detection_coco'


metainfo = {
    'classes': ('cola', 'pepsi', 'sprite', 'fanta', 'spring', 'ice', 'scream', 'milk', 'red', 'king'),
    'palette': [
        (255, 0, 0),    
        (0, 255, 0),    
        (0, 0, 255),    
        (255, 255, 0),  
        (255, 165, 0), 
        (128, 0, 128),  
        (255, 192, 203),  
        (255, 255, 255),  
        (255, 0, 0),    
        (0, 0, 0),     
    ]
}

num_classes = 10


max_epochs = 40

train_batch_size_per_gpu = 12

train_num_workers = 4


val_batch_size_per_gpu = 1
val_num_workers = 2


num_epochs_stage2 = 5


base_lr = 12 * 0.004 / (32*8)


load_from = 'https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_tiny_8xb32-300e_coco/rtmdet_tiny_8xb32-300e_coco_20220902_112414-78e30dcc.pth'  
model = dict(
    backbone=dict(frozen_stages=4),
    bbox_head=dict(dict(num_classes=num_classes)))

train_dataloader = dict(
    batch_size=train_batch_size_per_gpu,
    num_workers=train_num_workers,
    pin_memory=False,
    dataset=dict(
        data_root=data_root,
        metainfo=metainfo,
        ann_file='E:\github\MyOpenMMLab\homework_3_mmdetection\data\Drink_284_Detection_coco\\train_coco.json',
        data_prefix=dict(img='images/')))

val_dataloader = dict(
    batch_size=val_batch_size_per_gpu,
    num_workers=val_num_workers,
    dataset=dict(
        metainfo=metainfo,
        data_root=data_root,
        ann_file='E:\github\MyOpenMMLab\homework_3_mmdetection\data\Drink_284_Detection_coco\\val_coco.json',
        data_prefix=dict(img='images/')))

test_dataloader = val_dataloader

param_scheduler = [
    dict(
        type='LinearLR',
        start_factor=1.0e-5,
        by_epoch=False,
        begin=0,
        end=30),
    dict(
        type='CosineAnnealingLR',
        eta_min=base_lr * 0.05,
        begin=max_epochs // 2,  
        end=max_epochs,
        T_max=max_epochs // 2,
        by_epoch=True,
        convert_to_iter_based=True),
]
optim_wrapper = dict(optimizer=dict(lr=base_lr))


_base_.custom_hooks[1].switch_epoch = max_epochs - num_epochs_stage2

val_evaluator = dict(ann_file=data_root + '\\val_coco.json')
test_evaluator = val_evaluator

# һЩ��ӡ�����޸�
default_hooks = dict(
    checkpoint=dict(interval=10, max_keep_ckpts=2, save_best='auto'), 
    logger=dict(type='LoggerHook', interval=5))
train_cfg = dict(max_epochs=max_epochs, val_interval=10)

模型测试和推理

若想要进行离线测试，则可以使用提供的test.py脚本
模型特征图的可视化
Grad-Based CAM可视化
由于目标检测的特殊性，实际上使用Grad Box AM进行可视化
目标检测新趋势
随着大语言模型的飞速发展，目标检测逐渐从传统的封闭集合检测向和自然语言结合的开放类别检测发展，典型的方向如下：
1.Open-Vocabulary Object Detection：开放词汇目标检测，给定图片和类别词汇表，检测所有物体
2.Grounding Object Detection：给定图片和文本描述，预测文本中所提到的在图片中的物体位置
该方向的典型算法是GLIP，目前在MMDetection的dev-3.x中已经支持