diff --git a/docs/tutorials/config.md b/docs/tutorials/config.md index 59bcfb8647..0640759f4b 100644 --- a/docs/tutorials/config.md +++ b/docs/tutorials/config.md @@ -17,7 +17,7 @@ For example, if some modification is made based on PointPillars, user may first If you are building an entirely new method that does not share the structure with any of the existing methods, you may create a folder `xxx_rcnn` under `configs`, -Please refer to [mmcv](https://mmcv.readthedocs.io/en/latest/utils.html#config) for detailed documentation. +Please refer to [mmcv](https://mmcv.readthedocs.io/en/latest/understand_mmcv/config.html) for detailed documentation. ## Config Name Style @@ -116,20 +116,20 @@ model = dict( [0.50867593, 0.50656086, 0.30136237], [1.1511526, 1.0546296, 0.49706793], [0.47535285, 0.49249494, 0.5802117]]), # Mean sizes for each class, the order is consistent with class_names. - vote_moudule_cfg=dict( # Config to vote module branch, refer to mmdet3d.models.model_utils for more details + vote_moudule_cfg=dict( # Config of vote module branch, refer to mmdet3d.models.model_utils for more details in_channels=256, # Input channels for vote_module vote_per_seed=1, # Number of votes to generate for each seed gt_per_seed=3, # Number of gts for each seed conv_channels=(256, 256), # Channels for convolution - conv_cfg=dict(type='Conv1d'), # Config to convolution - norm_cfg=dict(type='BN1d'), # Config to normalization + conv_cfg=dict(type='Conv1d'), # Config of convolution + norm_cfg=dict(type='BN1d'), # Config of normalization norm_feats=True, # Whether to normalize features - vote_loss=dict( # Config to the loss function for voting branch + vote_loss=dict( # Config of the loss function for voting branch type='ChamferDistance', # Type of loss for voting branch mode='l1', # Loss mode of voting branch reduction='none', # Specifies the reduction to apply to the output loss_dst_weight=10.0)), # Destination loss weight of the voting branch - vote_aggregation_cfg=dict( # Config to vote aggregation branch + vote_aggregation_cfg=dict( # Config of vote aggregation branch type='PointSAModule', # type of vote aggregation module num_point=256, # Number of points for the set abstraction layer in vote aggregation branch radius=0.3, # Radius for the set abstraction layer in vote aggregation branch @@ -138,44 +138,44 @@ model = dict( use_xyz=True, # Whether to use xyz normalize_xyz=True), # Whether to normalize xyz feat_channels=(128, 128), # Channels for feature convolution - conv_cfg=dict(type='Conv1d'), # Config to convolution - norm_cfg=dict(type='BN1d'), # Config to normalization - objectness_loss=dict( # Config to objectness loss + conv_cfg=dict(type='Conv1d'), # Config of convolution + norm_cfg=dict(type='BN1d'), # Config of normalization + objectness_loss=dict( # Config of objectness loss type='CrossEntropyLoss', # Type of loss class_weight=[0.2, 0.8], # Class weight of the objectness loss reduction='sum', # Specifies the reduction to apply to the output loss_weight=5.0), # Loss weight of the objectness loss - center_loss=dict( # Config to center loss + center_loss=dict( # Config of center loss type='ChamferDistance', # Type of loss mode='l2', # Loss mode of center loss reduction='sum', # Specifies the reduction to apply to the output loss_src_weight=10.0, # Source loss weight of the voting branch. loss_dst_weight=10.0), # Destination loss weight of the voting branch. - dir_class_loss=dict( # Config to direction classification loss + dir_class_loss=dict( # Config of direction classification loss type='CrossEntropyLoss', # Type of loss reduction='sum', # Specifies the reduction to apply to the output loss_weight=1.0), # Loss weight of the direction classification loss - dir_res_loss=dict( # Config to direction residual loss + dir_res_loss=dict( # Config of direction residual loss type='SmoothL1Loss', # Type of loss reduction='sum', # Specifies the reduction to apply to the output loss_weight=10.0), # Loss weight of the direction residual loss - size_class_loss=dict( # Config to size classification loss + size_class_loss=dict( # Config of size classification loss type='CrossEntropyLoss', # Type of loss reduction='sum', # Specifies the reduction to apply to the output loss_weight=1.0), # Loss weight of the size classification loss - size_res_loss=dict( # Config to size residual loss + size_res_loss=dict( # Config of size residual loss type='SmoothL1Loss', # Type of loss reduction='sum', # Specifies the reduction to apply to the output loss_weight=3.3333333333333335), # Loss weight of the size residual loss - semantic_loss=dict( # Config to semantic loss + semantic_loss=dict( # Config of semantic loss type='CrossEntropyLoss', # Type of loss reduction='sum', # Specifies the reduction to apply to the output loss_weight=1.0)), # Loss weight of the semantic loss - train_cfg = dict( # Config of training hyperparameters for votenet + train_cfg = dict( # Config of training hyperparameters for VoteNet pos_distance_thr=0.3, # distance >= threshold 0.3 will be taken as positive samples - neg_distance_thr=0.6, # distance < threshold 0.6 will be taken as positive samples + neg_distance_thr=0.6, # distance < threshold 0.6 will be taken as negative samples sample_mod='vote'), # Mode of the sampling method - test_cfg = dict( # Config of testing hyperparameters for votenet + test_cfg = dict( # Config of testing hyperparameters for VoteNet sample_mod='seed', # Mode of the sampling method nms_thr=0.25, # The threshold to be used during NMS score_thr=0.8, # Threshold to filter out boxes @@ -195,7 +195,7 @@ train_pipeline = [ # Training pipeline, refer to mmdet3d.datasets.pipelines for dict( type='LoadAnnotations3D', # Second pipeline to load annotations, refer to mmdet3d.datasets.pipelines.indoor_loading for more details with_bbox_3d=True, # Whether to load 3D boxes - with_label_3d=True, # Whether to load 3D labels + with_label_3d=True, # Whether to load 3D labels corresponding to each 3D box with_mask_3d=True, # Whether to load 3D instance masks with_seg_3d=True), # Whether to load 3D semantic masks dict( @@ -210,7 +210,7 @@ train_pipeline = [ # Training pipeline, refer to mmdet3d.datasets.pipelines for flip_ratio_xz=0.5), # Probability of being flipped along xz plane dict( type='IndoorGlobalRotScale', # Augmentation pipeline that rotate and scale points and 3d boxes, refer to mmdet3d.datasets.pipelines.indoor_augment for more details - shift_height=True, # Whether to use height + shift_height=True, # Whether the loaded points use `shift_height` attribute rot_range=[-0.027777777777777776, 0.027777777777777776], # Range of rotation scale_range=None), # Range of scale dict( @@ -261,7 +261,7 @@ eval_pipeline = [ # Pipeline used for evaluation or visualization, refer to mmd ] data = dict( samples_per_gpu=8, # Batch size of a single GPU - workers_per_gpu=4, # Worker to pre-fetch data for each single GPU + workers_per_gpu=4, # Number of workers to pre-fetch data for each single GPU train=dict( # Train dataset config type='RepeatDataset', # Wrapper of dataset, refer to https://github.com/open-mmlab/mmdetection/blob/master/mmdet/datasets/dataset_wrappers.py for details. times=5, # Repeat times @@ -310,7 +310,7 @@ data = dict( 'pts_semantic_mask', 'pts_instance_mask' ]) ], - filter_empty_gt=False, # Whether to filter ground empty truth boxes + filter_empty_gt=False, # Whether to filter empty ground truth boxes classes=('cabinet', 'bed', 'chair', 'sofa', 'table', 'door', 'window', 'bookshelf', 'picture', 'counter', 'desk', 'curtain', 'refrigerator', 'showercurtrain', 'toilet', @@ -383,7 +383,7 @@ evaluation = dict(pipeline=[ # Pipeline is passed by eval_pipeline created befo ]) lr = 0.008 # Learning rate of optimizers optimizer = dict( # Config used to build optimizer, support all the optimizers in PyTorch whose arguments are also the same as those in PyTorch - type='Adam', # Type of optimizers, # Type of optimizers, refer to https://github.com/open-mmlab/mmcv/blob/v1.3.7/mmcv/runner/optimizer/default_constructor.py#L12 for more details + type='Adam', # Type of optimizers, refer to https://github.com/open-mmlab/mmcv/blob/v1.3.7/mmcv/runner/optimizer/default_constructor.py#L12 for more details lr=0.008) # Learning rate of optimizers, see detail usages of the parameters in the documentaion of PyTorch optimizer_config = dict( # Config used to build the optimizer hook, refer to https://github.com/open-mmlab/mmcv/blob/v1.3.7/mmcv/runner/hooks/optimizer.py#L22 for implementation details. grad_clip=dict( # Config used to grad_clip @@ -393,19 +393,19 @@ lr_config = dict( # Learning rate scheduler config used to register LrUpdater h policy='step', # The policy of scheduler, also support CosineAnnealing, Cyclic, etc. Refer to details of supported LrUpdater from https://github.com/open-mmlab/mmcv/blob/v1.3.7/mmcv/runner/hooks/lr_updater.py#L9. warmup=None, # The warmup policy, also support `exp` and `constant`. step=[24, 32]) # Steps to decay the learning rate -checkpoint_config = dict( # Config to set the checkpoint hook, Refer to https://github.com/open-mmlab/mmcv/blob/master/mmcv/runner/hooks/checkpoint.py for implementation. +checkpoint_config = dict( # Config of set the checkpoint hook, Refer to https://github.com/open-mmlab/mmcv/blob/master/mmcv/runner/hooks/checkpoint.py for implementation. interval=1) # The save interval is 1 -log_config = dict( # config to register logger hook +log_config = dict( # config of register logger hook interval=50, # Interval to print the log hooks=[dict(type='TextLoggerHook'), dict(type='TensorboardLoggerHook')]) # The logger used to record the training process. -runner = dict(type='EpochBasedRunner', max_epochs=36) # Runner that runs the workflow in total max_epochs +runner = dict(type='EpochBasedRunner', max_epochs=36) # Runner that runs the `workflow` in total `max_epochs` dist_params = dict(backend='nccl') # Parameters to setup distributed training, the port can also be set. log_level = 'INFO' # The level of logging. find_unused_parameters = True # Whether to find unused parameters work_dir = None # Directory to save the model checkpoints and logs for the current experiments. load_from = None # load models as a pre-trained model from a given path. This will not resume training. -resume_from = None # Resume checkpoints from a given path, the training will be resumed from the epoch when the checkpoint's is saved. +resume_from = None # Resume checkpoints from a given path, the training will be resumed from the epoch when the checkpoint's is saved. The training state such as the epoch number and optimizer state will be restored. workflow = [('train', 1)] # Workflow for runner. [('train', 1)] means there is only one workflow and the workflow named 'train' is executed once. The workflow trains the model by 36 epochs according to the max_epochs. gpu_ids = range(0, 1) # ids of gpus ``` diff --git a/docs_zh-CN/tutorials/config.md b/docs_zh-CN/tutorials/config.md index 0432a9f7dc..84ec13f5ba 100644 --- a/docs_zh-CN/tutorials/config.md +++ b/docs_zh-CN/tutorials/config.md @@ -1 +1,527 @@ -# 教程 1: 配置文件 +# 教程 1: 学习配置文件 + +我们在配置文件中支持了继承和模块化来方便进行各种实验。 +如果需要检查配置文件，可以通过运行 `python tools/misc/print_config.py /PATH/TO/CONFIG` 来查看完整的配置。 +你也可以传入 `--options xxx.yyy=zzz` 参数来查看更新后的配置。 + +## 配置文件结构 + +在 `config/_base_` 文件夹下有 4 个基本组件类型，分别是：数据集 (dataset)，模型 (model)，训练策略 (schedule) 和运行时的默认设置 (default runtime)。 +通过从上述每个文件夹中选取一个组件进行组合，许多方法如 SECOND、PointPillars、PartA2 和 VoteNet 都能够很容易地构建出来。 +由 `_base_` 下的组件组成的配置，被我们称为 _原始配置 (primitive)_。 + +对于同一文件夹下的所有配置，推荐**只有一个**对应的 _原始配置_ 文件，所有其他的配置文件都应该继承自这个 _原始配置_ 文件，这样就能保证配置文件的最大继承深度为 3。 + +为了便于理解，我们建议贡献者继承现有方法。 +例如，如果在 PointPillars 的基础上做了一些修改，用户首先可以通过指定 `_base_ = ../pointpillars/hv_pointpillars_fpn_sbn-all_4x8_2x_nus-3d.py` 来继承基础的 PointPillars 结构，然后修改配置文件中的必要参数以完成继承。 + +如果你在构建一个与任何现有方法不共享结构的全新方法，可以在 `configs` 文件夹下创建一个新的例如 `xxx_rcnn` 文件夹。 + +更多细节请参考 [MMCV](https://mmcv.readthedocs.io/en/latest/understand_mmcv/config.html) 文档。 + +## 配置文件名称风格 + +我们遵循以下样式来命名配置文件，并建议贡献者遵循相同的风格。 + +``` +{model}_[model setting]_{backbone}_{neck}_[norm setting]_[misc]_[gpu x batch_per_gpu]_{schedule}_{dataset} +``` + +`{xxx}` 是被要求填写的字段而 `[yyy]` 是可选的。 + +- `{model}`：模型种类，例如 `hv_pointpillars` (Hard Voxelization PointPillars)、`VoteNet` 等。 +- `[model setting]`：某些模型的特殊设定。 +- `{backbone}`： 主干网络种类例如 `regnet-400mf`、`regnet-1.6gf` 等。 +- `{neck}`：模型颈部的种类包括 `fpn`、`secfpn` 等。 +- `[norm_setting]`：如无特殊声明，默认使用 `bn` (Batch Normalization)，其他类型可以有 `gn` (Group Normalization)、`sbn` (Synchronized Batch Normalization) 等。 + `gn-head`/`gn-neck` 表示 GN 仅应用于网络的头部或颈部，而 `gn-all` 表示 GN 用于整个模型，例如主干网络、颈部和头部。 +- `[misc]`：模型中各式各样的设置/插件，例如 `strong-aug` 意味着在训练过程中使用更强的数据增广策略。 +- `[batch_per_gpu x gpu]`：每个 GPU 的样本数和 GPU 数量，默认使用 `4x8`。 +- `{schedule}`：训练方案，选项是 `1x`、`2x`、`20e` 等。 + `1x` 和 `2x` 分别代表训练 12 和 24 轮。 + `20e` 在级联模型中使用，表示训练 20 轮。 + 对于 `1x`/`2x`，初始学习率在第 8/16 和第 11/22 轮衰减 10 倍；对于 `20e`，初始学习率在第 16 和第 19 轮衰减 10 倍。 +- `{dataset}`：数据集，例如 `nus-3d`、`kitti-3d`、`lyft-3d`、`scannet-3d`、`sunrgbd-3d` 等。 + 当某一数据集存在多种设定时，我们也标记下所使用的类别数量，例如 `kitti-3d-3class` 和 `kitti-3d-car` 分别意味着在 KITTI 的所有三类上和单独车这一类上进行训练。 + +## 弃用的 train_cfg/test_cfg + +遵循 MMDetection 的做法，我们在配置文件中弃用 `train_cfg` 和 `test_cfg`，请在模型配置中指定它们。 +原始的配置结构如下： + +```python +# 已经弃用的形式 +model = dict( + type=..., + ... +) +train_cfg=dict(...) +test_cfg=dict(...) +``` + +迁移后的配置结构如下： + +```python +# 推荐的形式 +model = dict( + type=..., + ... + train_cfg=dict(...), + test_cfg=dict(...), +) +``` + +## VoteNet 配置文件示例 + +```python +model = dict( + type='VoteNet', # 检测器的类型，更多细节请参考 mmdet3d.models.detectors + backbone=dict( + type='PointNet2SASSG', # 主干网络的类型，更多细节请参考 mmdet3d.models.backbones + in_channels=4, # 点云输入通道数 + num_points=(2048, 1024, 512, 256), # 每个 SA 模块采样的中心点的数量 + radius=(0.2, 0.4, 0.8, 1.2), # 每个 SA 层的半径 + num_samples=(64, 32, 16, 16), # 每个 SA 层聚集的点的数量 + sa_channels=((64, 64, 128), (128, 128, 256), (128, 128, 256), + (128, 128, 256)), # SA 模块中每个多层感知器的输出通道数 + fp_channels=((256, 256), (256, 256)), # FP 模块中每个多层感知器的输出通道数 + norm_cfg=dict(type='BN2d'), # 归一化层的配置 + sa_cfg=dict( # 点集抽象 (SA) 模块的配置 + type='PointSAModule', # SA 模块的类型 + pool_mod='max', # SA 模块的池化方法 (最大池化或平均池化) + use_xyz=True, # 在特征聚合中是否使用 xyz 坐标 + normalize_xyz=True)), # 在特征聚合中是否使用标准化的 xyz 坐标 + bbox_head=dict( + type='VoteHead', # 检测框头的类型，更多细节请参考 mmdet3d.models.dense_heads + num_classes=18, # 分类的类别数量 + bbox_coder=dict( + type='PartialBinBasedBBoxCoder', # 框编码层的类型，更多细节请参考 mmdet3d.core.bbox.coders + num_sizes=18, # 尺寸聚类的数量 + num_dir_bins=1, # 编码方向角的间隔数 + with_rot=False, # 框是否带有旋转角度 + mean_sizes=[[0.76966727, 0.8116021, 0.92573744], + [1.876858, 1.8425595, 1.1931566], + [0.61328, 0.6148609, 0.7182701], + [1.3955007, 1.5121545, 0.83443564], + [0.97949594, 1.0675149, 0.6329687], + [0.531663, 0.5955577, 1.7500148], + [0.9624706, 0.72462326, 1.1481868], + [0.83221924, 1.0490936, 1.6875663], + [0.21132214, 0.4206159, 0.5372846], + [1.4440073, 1.8970833, 0.26985747], + [1.0294262, 1.4040797, 0.87554324], + [1.3766412, 0.65521795, 1.6813129], + [0.6650819, 0.71111923, 1.298853], + [0.41999173, 0.37906948, 1.7513971], + [0.59359556, 0.5912492, 0.73919016], + [0.50867593, 0.50656086, 0.30136237], + [1.1511526, 1.0546296, 0.49706793], + [0.47535285, 0.49249494, 0.5802117]]), # 每一类的平均尺寸，其顺序与类名顺序相同 + vote_moudule_cfg=dict( # 投票 (vote) 模块的配置，更多细节请参考 mmdet3d.models.model_utils + in_channels=256, # 投票模块的输入通道数 + vote_per_seed=1, # 对于每个种子点生成的投票数 + gt_per_seed=3, # 每个种子点的真实标签个数 + conv_channels=(256, 256), # 卷积通道数 + conv_cfg=dict(type='Conv1d'), # 卷积配置 + norm_cfg=dict(type='BN1d'), # 归一化层配置 + norm_feats=True, # 是否标准化特征 + vote_loss=dict( # 投票分支的损失函数配置 + type='ChamferDistance', # 投票分支的损失函数类型 + mode='l1', # 投票分支的损失函数模式 + reduction='none', # 设置对损失函数输出的聚合方法 + loss_dst_weight=10.0)), # 投票分支的目标损失权重 + vote_aggregation_cfg=dict( # 投票聚合分支的配置 + type='PointSAModule', # 投票聚合模块的类型 + num_point=256, # 投票聚合分支中 SA 模块的点的数量 + radius=0.3, # 投票聚合分支中 SA 模块的半径 + num_sample=16, # 投票聚合分支中 SA 模块的采样点的数量 + mlp_channels=[256, 128, 128, 128], # 投票聚合分支中 SA 模块的多层感知器的通道数 + use_xyz=True, # 是否使用 xyz 坐标 + normalize_xyz=True), # 是否使用标准化后的 xyz 坐标 + feat_channels=(128, 128), # 特征卷积的通道数 + conv_cfg=dict(type='Conv1d'), # 卷积的配置 + norm_cfg=dict(type='BN1d'), # 归一化层的配置 + objectness_loss=dict( # 物体性 (objectness) 损失函数的配置 + type='CrossEntropyLoss', # 损失函数类型 + class_weight=[0.2, 0.8], # 损失函数对每一类的权重 + reduction='sum', # 设置损失函数输出的聚合方法 + loss_weight=5.0), # 损失函数权重 + center_loss=dict( # 中心 (center) 损失函数的配置 + type='ChamferDistance', # 损失函数类型 + mode='l2', # 损失函数模式 + reduction='sum', # 设置损失函数输出的聚合方法 + loss_src_weight=10.0, # 源损失权重 + loss_dst_weight=10.0), # 目标损失权重 + dir_class_loss=dict( # 方向分类损失函数的配置 + type='CrossEntropyLoss', # 损失函数类型 + reduction='sum', # 设置损失函数输出的聚合方法 + loss_weight=1.0), # 损失函数权重 + dir_res_loss=dict( # 方向残差 (residual) 损失函数的配置 + type='SmoothL1Loss', # 损失函数类型 + reduction='sum', # 设置损失函数输出的聚合方法 + loss_weight=10.0), # 损失函数权重 + size_class_loss=dict( # 尺寸分类损失函数的配置 + type='CrossEntropyLoss', # 损失函数类型 + reduction='sum', # 设置损失函数输出的聚合方法 + loss_weight=1.0), # 损失函数权重 + size_res_loss=dict( # 尺寸残差损失函数的配置 + type='SmoothL1Loss', # 损失函数类型 + reduction='sum', # 设置损失函数输出的聚合方法 + loss_weight=3.3333333333333335), # 损失函数权重 + semantic_loss=dict( # 语义损失函数的配置 + type='CrossEntropyLoss', # 损失函数类型 + reduction='sum', # 设置损失函数输出的聚合方法 + loss_weight=1.0)), # 损失函数权重 + train_cfg = dict( # VoteNet 训练的超参数配置 + pos_distance_thr=0.3, # 距离 >= 0.3 阈值的样本将被视为正样本 + neg_distance_thr=0.6, # 距离 < 0.6 阈值的样本将被视为负样本 + sample_mod='vote'), # 采样方法的模式 + test_cfg = dict( # VoteNet 测试的超参数配置 + sample_mod='seed', # 采样方法的模式 + nms_thr=0.25, # NMS 中使用的阈值 + score_thr=0.8, # 剔除框的阈值 + per_class_proposal=False)) # 是否使用逐类提议框 (proposal) +dataset_type = 'ScanNetDataset' # 数据集类型 +data_root = './data/scannet/' # 数据路径 +class_names = ('cabinet', 'bed', 'chair', 'sofa', 'table', 'door', 'window', + 'bookshelf', 'picture', 'counter', 'desk', 'curtain', + 'refrigerator', 'showercurtrain', 'toilet', 'sink', 'bathtub', + 'garbagebin') # 类的名称 +train_pipeline = [ # 训练流水线，更多细节请参考 mmdet3d.datasets.pipelines + dict( + type='LoadPointsFromFile', # 第一个流程，用于读取点，更多细节请参考 mmdet3d.datasets.pipelines.indoor_loading + shift_height=True, # 是否使用变换高度 + load_dim=6, # 读取的点的维度 + use_dim=[0, 1, 2]), # 使用所读取点的哪些维度 + dict( + type='LoadAnnotations3D', # 第二个流程，用于读取标注，更多细节请参考 mmdet3d.datasets.pipelines.indoor_loading + with_bbox_3d=True, # 是否读取 3D 框 + with_label_3d=True, # 是否读取 3D 框对应的类别标签 + with_mask_3d=True, # 是否读取 3D 实例分割掩码 + with_seg_3d=True), # 是否读取 3D 语义分割掩码 + dict( + type='PointSegClassMapping', # 选取有效的类别，更多细节请参考 mmdet3d.datasets.pipelines.point_seg_class_mapping + valid_cat_ids=(3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 14, 16, 24, 28, 33, 34, + 36, 39), # 所有有效类别的编号 + max_cat_id=40), # 输入语义分割掩码中可能存在的最大类别编号 + dict(type='IndoorPointSample', # 室内点采样，更多细节请参考 mmdet3d.datasets.pipelines.indoor_sample + num_points=40000), # 采样的点的数量 + dict(type='IndoorFlipData', # 数据增广流程，随机翻转点和 3D 框 + flip_ratio_yz=0.5, # 沿着 yz 平面被翻转的概率 + flip_ratio_xz=0.5), # 沿着 xz 平面被翻转的概率 + dict( + type='IndoorGlobalRotScale', # 数据增广流程，旋转并放缩点和 3D 框，更多细节请参考 mmdet3d.datasets.pipelines.indoor_augment + shift_height=True, # 读取的点是否有高度这一属性 + rot_range=[-0.027777777777777776, 0.027777777777777776], # 旋转角范围 + scale_range=None), # 缩放尺寸范围 + dict( + type='DefaultFormatBundle3D', # 默认格式打包以收集读取的所有数据，更多细节请参考 mmdet3d.datasets.pipelines.formating + class_names=('cabinet', 'bed', 'chair', 'sofa', 'table', 'door', + 'window', 'bookshelf', 'picture', 'counter', 'desk', + 'curtain', 'refrigerator', 'showercurtrain', 'toilet', + 'sink', 'bathtub', 'garbagebin')), + dict( + type='Collect3D', # 最后一个流程，决定哪些键值对应的数据会被输入给检测器，更多细节请参考 mmdet3d.datasets.pipelines.formating + keys=[ + 'points', 'gt_bboxes_3d', 'gt_labels_3d', 'pts_semantic_mask', + 'pts_instance_mask' + ]) +] +test_pipeline = [ # 测试流水线，更多细节请参考 mmdet3d.datasets.pipelines + dict( + type='LoadPointsFromFile', # 第一个流程，用于读取点，更多细节请参考 mmdet3d.datasets.pipelines.indoor_loading + shift_height=True, # 是否使用变换高度 + load_dim=6, # 读取的点的维度 + use_dim=[0, 1, 2]), # 使用所读取点的哪些维度 + dict(type='IndoorPointSample', # 室内点采样，更多细节请参考 mmdet3d.datasets.pipelines.indoor_sample + num_points=40000), # 采样的点的数量 + dict( + type='DefaultFormatBundle3D', # 默认格式打包以收集读取的所有数据，更多细节请参考 mmdet3d.datasets.pipelines.formating + class_names=('cabinet', 'bed', 'chair', 'sofa', 'table', 'door', + 'window', 'bookshelf', 'picture', 'counter', 'desk', + 'curtain', 'refrigerator', 'showercurtrain', 'toilet', + 'sink', 'bathtub', 'garbagebin')), + dict(type='Collect3D', # 最后一个流程，决定哪些键值对应的数据会被输入给检测器，更多细节请参考 mmdet3d.datasets.pipelines.formating + keys=['points']) +] +eval_pipeline = [ # 模型验证或可视化所使用的流水线，更多细节请参考 mmdet3d.datasets.pipelines + dict( + type='LoadPointsFromFile', # 第一个流程，用于读取点，更多细节请参考 mmdet3d.datasets.pipelines.indoor_loading + shift_height=True, # 是否使用变换高度 + load_dim=6, # 读取的点的维度 + use_dim=[0, 1, 2]), # 使用所读取点的哪些维度 + dict( + type='DefaultFormatBundle3D', # 默认格式打包以收集读取的所有数据，更多细节请参考 mmdet3d.datasets.pipelines.formating + class_names=('cabinet', 'bed', 'chair', 'sofa', 'table', 'door', + 'window', 'bookshelf', 'picture', 'counter', 'desk', + 'curtain', 'refrigerator', 'showercurtrain', 'toilet', + 'sink', 'bathtub', 'garbagebin')), + with_label=False), + dict(type='Collect3D', # 最后一个流程，决定哪些键值对应的数据会被输入给检测器，更多细节请参考 mmdet3d.datasets.pipelines.formating + keys=['points']) +] +data = dict( + samples_per_gpu=8, # 单张 GPU 上的样本数 + workers_per_gpu=4, # 每张 GPU 上用于读取数据的进程数 + train=dict( # 训练数据集配置 + type='RepeatDataset', # 数据集嵌套，更多细节请参考 https://github.com/open-mmlab/mmdetection/blob/master/mmdet/datasets/dataset_wrappers.py + times=5, # 重复次数 + dataset=dict( + type='ScanNetDataset', # 数据集类型 + data_root='./data/scannet/', # 数据路径 + ann_file='./data/scannet/scannet_infos_train.pkl', # 数据标注文件的路径 + pipeline=[ # 流水线，这里传入的就是上面创建的训练流水线变量 + dict( + type='LoadPointsFromFile', + shift_height=True, + load_dim=6, + use_dim=[0, 1, 2]), + dict( + type='LoadAnnotations3D', + with_bbox_3d=True, + with_label_3d=True, + with_mask_3d=True, + with_seg_3d=True), + dict( + type='PointSegClassMapping', + valid_cat_ids=(3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 14, 16, 24, + 28, 33, 34, 36, 39), + max_cat_id=40), + dict(type='IndoorPointSample', num_points=40000), + dict( + type='IndoorFlipData', + flip_ratio_yz=0.5, + flip_ratio_xz=0.5), + dict( + type='IndoorGlobalRotScale', + shift_height=True, + rot_range=[-0.027777777777777776, 0.027777777777777776], + scale_range=None), + dict( + type='DefaultFormatBundle3D', + class_names=('cabinet', 'bed', 'chair', 'sofa', 'table', + 'door', 'window', 'bookshelf', 'picture', + 'counter', 'desk', 'curtain', 'refrigerator', + 'showercurtrain', 'toilet', 'sink', 'bathtub', + 'garbagebin')), + dict( + type='Collect3D', + keys=[ + 'points', 'gt_bboxes_3d', 'gt_labels_3d', + 'pts_semantic_mask', 'pts_instance_mask' + ]) + ], + filter_empty_gt=False, # 是否过滤掉空的标签框 + classes=('cabinet', 'bed', 'chair', 'sofa', 'table', 'door', + 'window', 'bookshelf', 'picture', 'counter', 'desk', + 'curtain', 'refrigerator', 'showercurtrain', 'toilet', + 'sink', 'bathtub', 'garbagebin'))), # 类别名称 + val=dict( # 验证数据集配置 + type='ScanNetDataset', # 数据集类型 + data_root='./data/scannet/', # 数据路径 + ann_file='./data/scannet/scannet_infos_val.pkl', # 数据标注文件的路径 + pipeline=[ # 流水线，这里传入的就是上面创建的测试流水线变量 + dict( + type='LoadPointsFromFile', + shift_height=True, + load_dim=6, + use_dim=[0, 1, 2]), + dict(type='IndoorPointSample', num_points=40000), + dict( + type='DefaultFormatBundle3D', + class_names=('cabinet', 'bed', 'chair', 'sofa', 'table', + 'door', 'window', 'bookshelf', 'picture', + 'counter', 'desk', 'curtain', 'refrigerator', + 'showercurtrain', 'toilet', 'sink', 'bathtub', + 'garbagebin')), + dict(type='Collect3D', keys=['points']) + ], + classes=('cabinet', 'bed', 'chair', 'sofa', 'table', 'door', 'window', + 'bookshelf', 'picture', 'counter', 'desk', 'curtain', + 'refrigerator', 'showercurtrain', 'toilet', 'sink', 'bathtub', + 'garbagebin'), # 类别名称 + test_mode=True), # 是否开启测试模式 + test=dict( # 测试数据集配置 + type='ScanNetDataset', # 数据集类型 + data_root='./data/scannet/', # 数据路径 + ann_file='./data/scannet/scannet_infos_val.pkl', # 数据标注文件的路径 + pipeline=[ # 流水线，这里传入的就是上面创建的测试流水线变量 + dict( + type='LoadPointsFromFile', + shift_height=True, + load_dim=6, + use_dim=[0, 1, 2]), + dict(type='IndoorPointSample', num_points=40000), + dict( + type='DefaultFormatBundle3D', + class_names=('cabinet', 'bed', 'chair', 'sofa', 'table', + 'door', 'window', 'bookshelf', 'picture', + 'counter', 'desk', 'curtain', 'refrigerator', + 'showercurtrain', 'toilet', 'sink', 'bathtub', + 'garbagebin')), + dict(type='Collect3D', keys=['points']) + ], + classes=('cabinet', 'bed', 'chair', 'sofa', 'table', 'door', 'window', + 'bookshelf', 'picture', 'counter', 'desk', 'curtain', + 'refrigerator', 'showercurtrain', 'toilet', 'sink', 'bathtub', + 'garbagebin'), # 类别名称 + test_mode=True)) # 是否开启测试模式 +evaluation = dict(pipeline=[ # 流水线，这里传入的就是上面创建的验证流水线变量 + dict( + type='LoadPointsFromFile', + coord_type='DEPTH', + shift_height=False, + load_dim=6, + use_dim=[0, 1, 2]), + dict( + type='DefaultFormatBundle3D', + class_names=('cabinet', 'bed', 'chair', 'sofa', 'table', 'door', + 'window', 'bookshelf', 'picture', 'counter', 'desk', + 'curtain', 'refrigerator', 'showercurtrain', 'toilet', + 'sink', 'bathtub', 'garbagebin'), + with_label=False), + dict(type='Collect3D', keys=['points']) +]) +lr = 0.008 # 优化器的学习率 +optimizer = dict( # 构建优化器所使用的配置，我们支持所有 PyTorch 中支持的优化器，并且拥有相同的参数名称 + type='Adam', # 优化器类型，更多细节请参考 https://github.com/open-mmlab/mmcv/blob/v1.3.7/mmcv/runner/optimizer/default_constructor.py#L12 + lr=0.008) # 优化器的学习率，用户可以在 PyTorch 文档中查看这些参数的详细使用方法 +optimizer_config = dict( # 构建优化器钩子的配置，更多实现细节可参考 https://github.com/open-mmlab/mmcv/blob/v1.3.7/mmcv/runner/hooks/optimizer.py#L22 + grad_clip=dict( # 梯度裁剪的配置 + max_norm=10, # 梯度的最大模长 + norm_type=2)) # 所使用的 p-范数的类型，可以设置成 'inf' 则指代无穷范数 +lr_config = dict( # 学习率策略配置，用于注册学习率更新的钩子 + policy='step', # 学习率调整的策略，支持 CosineAnnealing、Cyclic 等，更多支持的种类请参考 https://github.com/open-mmlab/mmcv/blob/v1.3.7/mmcv/runner/hooks/lr_updater.py#L9 + warmup=None, # Warmup 策略，同时也支持 `exp` 和 `constant` + step=[24, 32]) # 学习率衰减的步数 +checkpoint_config = dict( # 设置保存模型权重钩子的配置，具体实现请参考 https://github.com/open-mmlab/mmcv/blob/master/mmcv/runner/hooks/checkpoint.py + interval=1) # 保存模型权重的间隔是 1 轮 +log_config = dict( # 用于注册输出记录信息钩子的配置 + interval=50, # 输出记录信息的间隔 + hooks=[dict(type='TextLoggerHook'), + dict(type='TensorboardLoggerHook')]) # 用于记录训练过程的信息记录机制 +runner = dict(type='EpochBasedRunner', max_epochs=36) # 程序运行器，将会运行 `workflow` `max_epochs` 次 +dist_params = dict(backend='nccl') # 设置分布式训练的配置，通讯端口值也可被设置 +log_level = 'INFO' # 输出记录信息的等级 +find_unused_parameters = True # 是否查找模型中未使用的参数 +work_dir = None # 当前实验存储模型权重和输出信息的路径 +load_from = None # 从指定路径读取一个预训练的模型权重，这将不会继续 (resume) 训练 +resume_from = None # 从一个指定路径读入模型权重并继续训练，这意味着训练轮数、优化器状态等都将被读取 +workflow = [('train', 1)] # 要运行的工作流。[('train', 1)] 意味着只有一个名为 'train' 的工作流，它只会被执行一次。这一工作流依据 `max_epochs` 的值将会训练模型 36 轮。 +gpu_ids = range(0, 1) # 所使用的 GPU 编号 +``` + +## 常问问题 (FAQ) + +### 忽略基础配置文件里的部分内容 + +有时，您也许会需要通过设置 `_delete_=True` 来忽略基础配置文件里的一些域内容。 +请参照 [mmcv](https://mmcv.readthedocs.io/en/latest/understand_mmcv/config.html#inherit-from-base-config-with-ignored-fields) 来获得一些简单的指导。 + +例如在 MMDetection3D 中，为了改变如下所示 PointPillars FPN 模块的某些配置： + +```python +model = dict( + type='MVXFasterRCNN', + pts_voxel_layer=dict(...), + pts_voxel_encoder=dict(...), + pts_middle_encoder=dict(...), + pts_backbone=dict(...), + pts_neck=dict( + type='FPN', + norm_cfg=dict(type='naiveSyncBN2d', eps=1e-3, momentum=0.01), + act_cfg=dict(type='ReLU'), + in_channels=[64, 128, 256], + out_channels=256, + start_level=0, + num_outs=3), + pts_bbox_head=dict(...)) +``` + +`FPN` 和 `SECONDFPN` 使用不同的关键词来构建。 + +```python +_base_ = '../_base_/models/hv_pointpillars_fpn_nus.py' +model = dict( + pts_neck=dict( + _delete_=True, + type='SECONDFPN', + norm_cfg=dict(type='naiveSyncBN2d', eps=1e-3, momentum=0.01), + in_channels=[64, 128, 256], + upsample_strides=[1, 2, 4], + out_channels=[128, 128, 128]), + pts_bbox_head=dict(...)) +``` + +`_delete_=True` 的标识将会使用新的键值覆盖掉 `pts_neck` 中的所有旧键值。 + +### 使用配置文件里的中间变量 + +配置文件里会使用一些中间变量，例如数据集中的 `train_pipeline`/`test_pipeline`。 +值得注意的是，当修改子配置文件中的中间变量后，用户还需再次将其传入相应字段。 +例如，我们想在训练和测试中，对 PointPillars 使用多尺度策略 (multi scale strategy)，那么 `train_pipeline`/`test_pipeline` 就是我们想要修改的中间变量。 + +```python +_base_ = './nus-3d.py' +train_pipeline = [ + dict( + type='LoadPointsFromFile', + load_dim=5, + use_dim=5, + file_client_args=file_client_args), + dict( + type='LoadPointsFromMultiSweeps', + sweeps_num=10, + file_client_args=file_client_args), + dict(type='LoadAnnotations3D', with_bbox_3d=True, with_label_3d=True), + dict( + type='GlobalRotScaleTrans', + rot_range=[-0.3925, 0.3925], + scale_ratio_range=[0.95, 1.05], + translation_std=[0, 0, 0]), + dict(type='RandomFlip3D', flip_ratio_bev_horizontal=0.5), + dict(type='PointsRangeFilter', point_cloud_range=point_cloud_range), + dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range), + dict(type='ObjectNameFilter', classes=class_names), + dict(type='PointShuffle'), + dict(type='DefaultFormatBundle3D', class_names=class_names), + dict(type='Collect3D', keys=['points', 'gt_bboxes_3d', 'gt_labels_3d']) +] +test_pipeline = [ + dict( + type='LoadPointsFromFile', + load_dim=5, + use_dim=5, + file_client_args=file_client_args), + dict( + type='LoadPointsFromMultiSweeps', + sweeps_num=10, + file_client_args=file_client_args), + dict( + type='MultiScaleFlipAug3D', + img_scale=(1333, 800), + pts_scale_ratio=[0.95, 1.0, 1.05], + flip=False, + transforms=[ + dict( + type='GlobalRotScaleTrans', + rot_range=[0, 0], + scale_ratio_range=[1., 1.], + translation_std=[0, 0, 0]), + dict(type='RandomFlip3D'), + dict( + type='PointsRangeFilter', point_cloud_range=point_cloud_range), + dict( + type='DefaultFormatBundle3D', + class_names=class_names, + with_label=False), + dict(type='Collect3D', keys=['points']) + ]) +] +data = dict( + train=dict(pipeline=train_pipeline), + val=dict(pipeline=test_pipeline), + test=dict(pipeline=test_pipeline)) +``` + +这里，我们首先定义了新的 `train_pipeline`/`test_pipeline`，然后将其传入 `data`。