mmdetection3d-之（一）–FCOS3d训练nuscenes-mini数据集

参考网上的博客，出现各种错误，最大的是：

AssertionError: Samples in split doesn’t match samples in predictions.

给了解决方案，也不知道那个数字是怎么来的。索性自己来一遍，参考了github issue。 

 第一步，下载数据集并解压：

wget https://www.nuscenes.org/data/v1.0-mini.tgz  # Download the nuScenes mini split.
tar -xf v1.0-mini.tgz -C /data/sets/nuscenes  # Uncompress the nuScenes mini split.

第二步，修改代码 tools/create_data.py 

    elif args.dataset == 'nuscenes' and args.version != 'v1.0-mini':
        train_version = f'{args.version}-trainval'
        nuscenes_data_prep(
            root_path=args.root_path,
            info_prefix=args.extra_tag,
            version=train_version,
            dataset_name='NuScenesDataset',
            out_dir=args.out_dir,
            max_sweeps=args.max_sweeps)

        """
        test_version = f'{args.version}-test'
        nuscenes_data_prep(
            root_path=args.root_path,
            info_prefix=args.extra_tag,
            version=test_version,
            dataset_name='NuScenesDataset',
            out_dir=args.out_dir,
            max_sweeps=args.max_sweeps) 
        """

 第三步，制作数据集

# 进入解压目录
cd /data/sets/nuscenes
# split重命名
mv v1.0-mini v1.0-trainval
python tools/create_data.py nuscenes（固定） --root-path ./data/nuscenes --out-dir ./data/nuscenes --extra-tag nuscenes（固定） --workers 20

 第四步，修改配置文件

configs/_base_/datasets/nus-mini-mono3d.py

dataset_type = 'NuScenesDataset'
data_root = 'data/nuscenes/'

configs/fcos3d/fcos3d_r101-caffe-dcn_fpn_head-gn_8xb2-1x_nus-mini-mono3d.py

_base_ = [
    '../_base_/datasets/nus-mini-mono3d.py', '../_base_/models/fcos3d.py',
    '../_base_/schedules/mmdet-schedule-1x.py', '../_base_/default_runtime.py'
]

...
#batch_size 根据需要修改
train_dataloader = dict(
    batch_size=4, num_workers=4, dataset=dict(pipeline=train_pipeline))
test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
val_dataloader = dict(dataset=dict(pipeline=test_pipeline))

...

# 修改训练epoch
# learning rate
param_scheduler = [
    dict(
        type='LinearLR',
        start_factor=1.0 / 3,
        by_epoch=False,
        begin=0,
        end=200),
    dict(
        type='MultiStepLR',
        begin=0,
        end=12,
        by_epoch=True,
        milestones=[8, 11],
        gamma=0.1)
]

第五步，多卡单卡训练

# 多卡训练
CUDA_VISIBLE_DEVICES=0,1 tools/dist_train.sh configs/fcos3d/fcos3d_r101-caffe-dcn_fpn_head-gn_8xb2-1x_nus-mini-mono3d.py 2

# 单卡训练
CUDA_VISIBLE_DEVICES=0 tools/dist_train.sh configs/fcos3d/fcos3d_r101-caffe-dcn_fpn_head-gn_8xb2-1x_nus-mini-mono3d.py 1