将子模块转换为普通目录

docker-hub/BEVFormer/BEVFormer/tools/dist_train_numa.sh

+#!/usr/bin/env bash
+
+MASTER_ADDR=${1:-localhost}
+MASTER_PORT=6000
+NNODES=${2:-1}
+NODE_RANK=${3:-0}
+GPUS_PER_NODE=8
+DISTRIBUTED_ARGS="--nproc_per_node $GPUS_PER_NODE --nnodes $NNODES --node_rank $NODE_RANK --master_addr $MASTER_ADDR --master_port $MASTER_PORT"
+
+CONFIG=$4
+
+# add numa affinity config
+PYTHONPATH="$(dirname $0)/..":$PYTHONPATH \
+torchrun $DISTRIBUTED_ARGS --no-python \
+    bash -c '
+    numa_map=( $(hy-smi --showtopo | grep "Numa Node" | awk "{print \$6}") );
+    LOCAL_RANK=${LOCAL_RANK:-0}
+    NUMA_ID=${numa_map[$LOCAL_RANK]}
+    numactl --cpunodebind=${NUMA_ID} --membind=${NUMA_ID} python $(dirname "$0")/tools/train.py "$@"
+    ' _ $CONFIG --launcher pytorch ${@:5} --deterministic

docker-hub/BEVFormer/BEVFormer/tools/fp16/dist_train.sh

+#!/usr/bin/env bash
+
+CONFIG=$1
+GPUS=$2
+PORT=${PORT:-28508}
+
+PYTHONPATH="$(dirname $0)/..":$PYTHONPATH \
+python -m torch.distributed.launch --nproc_per_node=$GPUS --master_port=$PORT \
+    $(dirname "$0")/train.py $CONFIG --launcher pytorch ${@:3} --deterministic

docker-hub/BEVFormer/BEVFormer/tools/fp16/train.py

+# Copyright (c) OpenMMLab. All rights reserved.
+from __future__ import division
+
+import argparse
+import copy
+import mmcv
+import os
+import time
+import torch
+import warnings
+from mmcv import Config, DictAction
+from mmcv.runner import get_dist_info, init_dist, wrap_fp16_model
+from os import path as osp
+
+from mmdet import __version__ as mmdet_version
+from mmdet3d import __version__ as mmdet3d_version
+#from mmdet3d.apis import train_model
+
+from mmdet3d.datasets import build_dataset
+from mmdet3d.models import build_model
+from mmdet3d.utils import collect_env, get_root_logger
+from mmdet.apis import set_random_seed
+from mmseg import __version__ as mmseg_version
+
+from mmcv.utils import TORCH_VERSION, digit_version
+
+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(
+        '--no-validate',
+        action='store_true',
+        help='whether not to evaluate the checkpoint during training')
+    group_gpus = parser.add_mutually_exclusive_group()
+    group_gpus.add_argument(
+        '--gpus',
+        type=int,
+        help='number of gpus to use '
+        '(only applicable to non-distributed training)')
+    group_gpus.add_argument(
+        '--gpu-ids',
+        type=int,
+        nargs='+',
+        help='ids of gpus to use '
+        '(only applicable to non-distributed training)')
+    parser.add_argument('--seed', type=int, default=0, help='random seed')
+    parser.add_argument(
+        '--deterministic',
+        action='store_true',
+        help='whether to set deterministic options for CUDNN backend.')
+    parser.add_argument(
+        '--options',
+        nargs='+',
+        action=DictAction,
+        help='override some settings in the used config, the key-value pair '
+        'in xxx=yyy format will be merged into config file (deprecate), '
+        'change to --cfg-options instead.')
+    parser.add_argument(
+        '--cfg-options',
+        nargs='+',
+        action=DictAction,
+        help='override some settings in the used config, the key-value pair '
+        'in xxx=yyy format will be merged into config file. If the value to '
+        'be overwritten is a list, it should be like key="[a,b]" or key=a,b '
+        'It also allows nested list/tuple values, e.g. key="[(a,b),(c,d)]" '
+        'Note that the quotation marks are necessary and that no white space '
+        'is allowed.')
+    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)
+
+    if args.options and args.cfg_options:
+        raise ValueError(
+            '--options and --cfg-options cannot be both specified, '
+            '--options is deprecated in favor of --cfg-options')
+    if args.options:
+        warnings.warn('--options is deprecated in favor of --cfg-options')
+        args.cfg_options = args.options
+
+    return args
+
+
+def main():
+    args = parse_args()
+
+    cfg = Config.fromfile(args.config)
+    if args.cfg_options is not None:
+        cfg.merge_from_dict(args.cfg_options)
+    # import modules from string list.
+    if cfg.get('custom_imports', None):
+        from mmcv.utils import import_modules_from_strings
+        import_modules_from_strings(**cfg['custom_imports'])
+
+    # import modules from plguin/xx, registry will be updated
+    if hasattr(cfg, 'plugin'):
+        if cfg.plugin:
+            import importlib
+            if hasattr(cfg, 'plugin_dir'):
+                plugin_dir = cfg.plugin_dir
+                _module_dir = os.path.dirname(plugin_dir)
+                _module_dir = _module_dir.split('/')
+                _module_path = _module_dir[0]
+
+                for m in _module_dir[1:]:
+                    _module_path = _module_path + '.' + m
+                print(_module_path)
+                plg_lib = importlib.import_module(_module_path)
+            else:
+                # import dir is the dirpath for the config file
+                _module_dir = os.path.dirname(args.config)
+                _module_dir = _module_dir.split('/')
+                _module_path = _module_dir[0]
+                for m in _module_dir[1:]:
+                    _module_path = _module_path + '.' + m
+                print(_module_path)
+                plg_lib = importlib.import_module(_module_path)
+            
+            from projects.mmdet3d_plugin.bevformer.apis import custom_train_model
+    # set cudnn_benchmark
+    if cfg.get('cudnn_benchmark', False):
+        torch.backends.cudnn.benchmark = True
+
+    # work_dir is determined in this priority: CLI > segment in file > filename
+    if args.work_dir is not None:
+        # update configs according to CLI args if args.work_dir is not None
+        cfg.work_dir = args.work_dir
+    elif cfg.get('work_dir', None) is None:
+        # use config filename as default work_dir if cfg.work_dir is None
+        cfg.work_dir = osp.join('./work_dirs',
+                                osp.splitext(osp.basename(args.config))[0])
+    #if args.resume_from is not None:
+
+    if args.resume_from is not None and osp.isfile(args.resume_from): 
+        cfg.resume_from = args.resume_from
+
+    if args.gpu_ids is not None:
+        cfg.gpu_ids = args.gpu_ids
+    else:
+        cfg.gpu_ids = range(1) if args.gpus is None else range(args.gpus)
+    if digit_version(TORCH_VERSION) != digit_version('1.8.1'):
+        cfg.optimizer['type'] = 'AdamW'
+    if args.autoscale_lr:
+        # apply the linear scaling rule (https://arxiv.org/abs/1706.02677)
+        cfg.optimizer['lr'] = cfg.optimizer['lr'] * len(cfg.gpu_ids) / 8
+
+    # init distributed env first, since logger depends on the dist info.
+    if args.launcher == 'none':
+        assert False, 'DOT NOT SUPPORT!!!'
+        distributed = False
+    else:
+        distributed = True
+        init_dist(args.launcher, **cfg.dist_params)
+        # re-set gpu_ids with distributed training mode
+        _, world_size = get_dist_info()
+        cfg.gpu_ids = range(world_size)
+
+    # create work_dir
+    mmcv.mkdir_or_exist(osp.abspath(cfg.work_dir))
+    # dump config
+    cfg.dump(osp.join(cfg.work_dir, osp.basename(args.config)))
+    # init the logger before other steps
+    timestamp = time.strftime('%Y%m%d_%H%M%S', time.localtime())
+    log_file = osp.join(cfg.work_dir, f'{timestamp}.log')
+    # specify logger name, if we still use 'mmdet', the output info will be
+    # filtered and won't be saved in the log_file
+    # TODO: ugly workaround to judge whether we are training det or seg model
+    if cfg.model.type in ['EncoderDecoder3D']:
+        logger_name = 'mmseg'
+    else:
+        logger_name = 'mmdet'
+    logger = get_root_logger(
+        log_file=log_file, log_level=cfg.log_level, name=logger_name)
+
+    # 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([(f'{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
+    meta['config'] = cfg.pretty_text
+
+    # log some basic info
+    logger.info(f'Distributed training: {distributed}')
+    logger.info(f'Config:\n{cfg.pretty_text}')
+
+    # set random seeds
+    if args.seed is not None:
+        logger.info(f'Set random seed to {args.seed}, '
+                    f'deterministic: {args.deterministic}')
+        set_random_seed(args.seed, deterministic=args.deterministic)
+    cfg.seed = args.seed
+    meta['seed'] = args.seed
+    meta['exp_name'] = osp.basename(args.config)
+
+    model = build_model(
+        cfg.model,
+        train_cfg=cfg.get('train_cfg'),
+        test_cfg=cfg.get('test_cfg'))
+    model.init_weights()
+
+    eval_model_config = copy.deepcopy(cfg.model)
+    eval_model = build_model(
+        eval_model_config,
+        train_cfg=cfg.get('train_cfg'),
+        test_cfg=cfg.get('test_cfg'))
+    
+    fp16_cfg = cfg.get('fp16', None)
+    if fp16_cfg is not None:
+        wrap_fp16_model(eval_model)
+
+    #eval_model.init_weights()
+    eval_model.load_state_dict(model.state_dict())
+
+    logger.info(f'Model:\n{model}')
+    from projects.mmdet3d_plugin.datasets import custom_build_dataset
+    datasets = [custom_build_dataset(cfg.data.train)]
+    if len(cfg.workflow) == 2:
+        val_dataset = copy.deepcopy(cfg.data.val)
+        # in case we use a dataset wrapper
+        if 'dataset' in cfg.data.train:
+            val_dataset.pipeline = cfg.data.train.dataset.pipeline
+        else:
+            val_dataset.pipeline = cfg.data.train.pipeline
+        # set test_mode=False here in deep copied config
+        # which do not affect AP/AR calculation later
+        # refer to https://mmdetection3d.readthedocs.io/en/latest/tutorials/customize_runtime.html#customize-workflow  # noqa
+        val_dataset.test_mode = False
+        datasets.append(custom_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=mmdet_version,
+            mmseg_version=mmseg_version,
+            mmdet3d_version=mmdet3d_version,
+            config=cfg.pretty_text,
+            CLASSES=datasets[0].CLASSES,
+            PALETTE=datasets[0].PALETTE  # for segmentors
+            if hasattr(datasets[0], 'PALETTE') else None)
+    # add an attribute for visualization convenience
+    model.CLASSES = datasets[0].CLASSES
+    custom_train_model(
+        model,
+        datasets,
+        cfg,
+        eval_model=eval_model,
+        distributed=distributed,
+        validate=(not args.no_validate),
+        timestamp=timestamp,
+        meta=meta)
+
+
+if __name__ == '__main__':
+    main()

docker-hub/BEVFormer/BEVFormer/tools/misc/browse_dataset.py

+# Copyright (c) OpenMMLab. All rights reserved.
+import argparse
+import numpy as np
+import warnings
+from mmcv import Config, DictAction, mkdir_or_exist, track_iter_progress
+from os import path as osp
+
+from mmdet3d.core.bbox import (Box3DMode, CameraInstance3DBoxes, Coord3DMode,
+                               DepthInstance3DBoxes, LiDARInstance3DBoxes)
+from mmdet3d.core.visualizer import (show_multi_modality_result, show_result,
+                                     show_seg_result)
+from mmdet3d.datasets import build_dataset
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(description='Browse a dataset')
+    parser.add_argument('config', help='train config file path')
+    parser.add_argument(
+        '--skip-type',
+        type=str,
+        nargs='+',
+        default=['Normalize'],
+        help='skip some useless pipeline')
+    parser.add_argument(
+        '--output-dir',
+        default=None,
+        type=str,
+        help='If there is no display interface, you can save it')
+    parser.add_argument(
+        '--task',
+        type=str,
+        choices=['det', 'seg', 'multi_modality-det', 'mono-det'],
+        help='Determine the visualization method depending on the task.')
+    parser.add_argument(
+        '--online',
+        action='store_true',
+        help='Whether to perform online visualization. Note that you often '
+        'need a monitor to do so.')
+    parser.add_argument(
+        '--cfg-options',
+        nargs='+',
+        action=DictAction,
+        help='override some settings in the used config, the key-value pair '
+        'in xxx=yyy format will be merged into config file. If the value to '
+        'be overwritten is a list, it should be like key="[a,b]" or key=a,b '
+        'It also allows nested list/tuple values, e.g. key="[(a,b),(c,d)]" '
+        'Note that the quotation marks are necessary and that no white space '
+        'is allowed.')
+    args = parser.parse_args()
+    return args
+
+
+def build_data_cfg(config_path, skip_type, cfg_options):
+    """Build data config for loading visualization data."""
+    cfg = Config.fromfile(config_path)
+    if cfg_options is not None:
+        cfg.merge_from_dict(cfg_options)
+    # import modules from string list.
+    if cfg.get('custom_imports', None):
+        from mmcv.utils import import_modules_from_strings
+        import_modules_from_strings(**cfg['custom_imports'])
+    # extract inner dataset of `RepeatDataset` as `cfg.data.train`
+    # so we don't need to worry about it later
+    if cfg.data.train['type'] == 'RepeatDataset':
+        cfg.data.train = cfg.data.train.dataset
+    # use only first dataset for `ConcatDataset`
+    if cfg.data.train['type'] == 'ConcatDataset':
+        cfg.data.train = cfg.data.train.datasets[0]
+    train_data_cfg = cfg.data.train
+    # eval_pipeline purely consists of loading functions
+    # use eval_pipeline for data loading
+    train_data_cfg['pipeline'] = [
+        x for x in cfg.eval_pipeline if x['type'] not in skip_type
+    ]
+
+    return cfg
+
+
+def to_depth_mode(points, bboxes):
+    """Convert points and bboxes to Depth Coord and Depth Box mode."""
+    if points is not None:
+        points = Coord3DMode.convert_point(points.copy(), Coord3DMode.LIDAR,
+                                           Coord3DMode.DEPTH)
+    if bboxes is not None:
+        bboxes = Box3DMode.convert(bboxes.clone(), Box3DMode.LIDAR,
+                                   Box3DMode.DEPTH)
+    return points, bboxes
+
+
+def show_det_data(idx, dataset, out_dir, filename, show=False):
+    """Visualize 3D point cloud and 3D bboxes."""
+    example = dataset.prepare_train_data(idx)
+    points = example['points']._data.numpy()
+    gt_bboxes = dataset.get_ann_info(idx)['gt_bboxes_3d'].tensor
+    if dataset.box_mode_3d != Box3DMode.DEPTH:
+        points, gt_bboxes = to_depth_mode(points, gt_bboxes)
+    show_result(
+        points,
+        gt_bboxes.clone(),
+        None,
+        out_dir,
+        filename,
+        show=show,
+        snapshot=True)
+
+
+def show_seg_data(idx, dataset, out_dir, filename, show=False):
+    """Visualize 3D point cloud and segmentation mask."""
+    example = dataset.prepare_train_data(idx)
+    points = example['points']._data.numpy()
+    gt_seg = example['pts_semantic_mask']._data.numpy()
+    show_seg_result(
+        points,
+        gt_seg.copy(),
+        None,
+        out_dir,
+        filename,
+        np.array(dataset.PALETTE),
+        dataset.ignore_index,
+        show=show,
+        snapshot=True)
+
+
+def show_proj_bbox_img(idx,
+                       dataset,
+                       out_dir,
+                       filename,
+                       show=False,
+                       is_nus_mono=False):
+    """Visualize 3D bboxes on 2D image by projection."""
+    try:
+        example = dataset.prepare_train_data(idx)
+    except AttributeError:  # for Mono-3D datasets
+        example = dataset.prepare_train_img(idx)
+    gt_bboxes = dataset.get_ann_info(idx)['gt_bboxes_3d']
+    img_metas = example['img_metas']._data
+    img = example['img']._data.numpy()
+    # need to transpose channel to first dim
+    img = img.transpose(1, 2, 0)
+    # no 3D gt bboxes, just show img
+    if gt_bboxes.tensor.shape[0] == 0:
+        gt_bboxes = None
+    if isinstance(gt_bboxes, DepthInstance3DBoxes):
+        show_multi_modality_result(
+            img,
+            gt_bboxes,
+            None,
+            None,
+            out_dir,
+            filename,
+            box_mode='depth',
+            img_metas=img_metas,
+            show=show)
+    elif isinstance(gt_bboxes, LiDARInstance3DBoxes):
+        show_multi_modality_result(
+            img,
+            gt_bboxes,
+            None,
+            img_metas['lidar2img'],
+            out_dir,
+            filename,
+            box_mode='lidar',
+            img_metas=img_metas,
+            show=show)
+    elif isinstance(gt_bboxes, CameraInstance3DBoxes):
+        show_multi_modality_result(
+            img,
+            gt_bboxes,
+            None,
+            img_metas['cam2img'],
+            out_dir,
+            filename,
+            box_mode='camera',
+            img_metas=img_metas,
+            show=show)
+    else:
+        # can't project, just show img
+        warnings.warn(
+            f'unrecognized gt box type {type(gt_bboxes)}, only show image')
+        show_multi_modality_result(
+            img, None, None, None, out_dir, filename, show=show)
+
+
+def main():
+    args = parse_args()
+
+    if args.output_dir is not None:
+        mkdir_or_exist(args.output_dir)
+
+    cfg = build_data_cfg(args.config, args.skip_type, args.cfg_options)
+    try:
+        dataset = build_dataset(
+            cfg.data.train, default_args=dict(filter_empty_gt=False))
+    except TypeError:  # seg dataset doesn't have `filter_empty_gt` key
+        dataset = build_dataset(cfg.data.train)
+    data_infos = dataset.data_infos
+    dataset_type = cfg.dataset_type
+
+    # configure visualization mode
+    vis_task = args.task  # 'det', 'seg', 'multi_modality-det', 'mono-det'
+
+    for idx, data_info in enumerate(track_iter_progress(data_infos)):
+        if dataset_type in ['KittiDataset', 'WaymoDataset']:
+            data_path = data_info['point_cloud']['velodyne_path']
+        elif dataset_type in [
+                'ScanNetDataset', 'SUNRGBDDataset', 'ScanNetSegDataset',
+                'S3DISSegDataset', 'S3DISDataset'
+        ]:
+            data_path = data_info['pts_path']
+        elif dataset_type in ['NuScenesDataset', 'LyftDataset']:
+            data_path = data_info['lidar_path']
+        elif dataset_type in ['NuScenesMonoDataset']:
+            data_path = data_info['file_name']
+        else:
+            raise NotImplementedError(
+                f'unsupported dataset type {dataset_type}')
+
+        file_name = osp.splitext(osp.basename(data_path))[0]
+
+        if vis_task in ['det', 'multi_modality-det']:
+            # show 3D bboxes on 3D point clouds
+            show_det_data(
+                idx, dataset, args.output_dir, file_name, show=args.online)
+        if vis_task in ['multi_modality-det', 'mono-det']:
+            # project 3D bboxes to 2D image
+            show_proj_bbox_img(
+                idx,
+                dataset,
+                args.output_dir,
+                file_name,
+                show=args.online,
+                is_nus_mono=(dataset_type == 'NuScenesMonoDataset'))
+        elif vis_task in ['seg']:
+            # show 3D segmentation mask on 3D point clouds
+            show_seg_data(
+                idx, dataset, args.output_dir, file_name, show=args.online)
+
+
+if __name__ == '__main__':
+    main()

docker-hub/BEVFormer/BEVFormer/tools/misc/fuse_conv_bn.py

+# Copyright (c) OpenMMLab. All rights reserved.
+import argparse
+import torch
+from mmcv.runner import save_checkpoint
+from torch import nn as nn
+
+from mmdet.apis import init_model
+
+
+def fuse_conv_bn(conv, bn):
+    """During inference, the functionary of batch norm layers is turned off but
+    only the mean and var alone channels are used, which exposes the chance to
+    fuse it with the preceding conv layers to save computations and simplify
+    network structures."""
+    conv_w = conv.weight
+    conv_b = conv.bias if conv.bias is not None else torch.zeros_like(
+        bn.running_mean)
+
+    factor = bn.weight / torch.sqrt(bn.running_var + bn.eps)
+    conv.weight = nn.Parameter(conv_w *
+                               factor.reshape([conv.out_channels, 1, 1, 1]))
+    conv.bias = nn.Parameter((conv_b - bn.running_mean) * factor + bn.bias)
+    return conv
+
+
+def fuse_module(m):
+    last_conv = None
+    last_conv_name = None
+
+    for name, child in m.named_children():
+        if isinstance(child, (nn.BatchNorm2d, nn.SyncBatchNorm)):
+            if last_conv is None:  # only fuse BN that is after Conv
+                continue
+            fused_conv = fuse_conv_bn(last_conv, child)
+            m._modules[last_conv_name] = fused_conv
+            # To reduce changes, set BN as Identity instead of deleting it.
+            m._modules[name] = nn.Identity()
+            last_conv = None
+        elif isinstance(child, nn.Conv2d):
+            last_conv = child
+            last_conv_name = name
+        else:
+            fuse_module(child)
+    return m
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(
+        description='fuse Conv and BN layers in a model')
+    parser.add_argument('config', help='config file path')
+    parser.add_argument('checkpoint', help='checkpoint file path')
+    parser.add_argument('out', help='output path of the converted model')
+    args = parser.parse_args()
+    return args
+
+
+def main():
+    args = parse_args()
+    # build the model from a config file and a checkpoint file
+    model = init_model(args.config, args.checkpoint)
+    # fuse conv and bn layers of the model
+    fused_model = fuse_module(model)
+    save_checkpoint(fused_model, args.out)
+
+
+if __name__ == '__main__':
+    main()

docker-hub/BEVFormer/BEVFormer/tools/misc/print_config.py

+# Copyright (c) OpenMMLab. All rights reserved.
+import argparse
+from mmcv import Config, DictAction
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(description='Print the whole config')
+    parser.add_argument('config', help='config file path')
+    parser.add_argument(
+        '--options', nargs='+', action=DictAction, help='arguments in dict')
+    args = parser.parse_args()
+
+    return args
+
+
+def main():
+    args = parse_args()
+
+    cfg = Config.fromfile(args.config)
+    if args.options is not None:
+        cfg.merge_from_dict(args.options)
+    print(f'Config:\n{cfg.pretty_text}')
+
+
+if __name__ == '__main__':
+    main()

docker-hub/BEVFormer/BEVFormer/tools/misc/visualize_results.py

+# Copyright (c) OpenMMLab. All rights reserved.
+import argparse
+import mmcv
+from mmcv import Config
+
+from mmdet3d.datasets import build_dataset
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(
+        description='MMDet3D visualize the results')
+    parser.add_argument('config', help='test config file path')
+    parser.add_argument('--result', help='results file in pickle format')
+    parser.add_argument(
+        '--show-dir', help='directory where visualize results will be saved')
+    args = parser.parse_args()
+
+    return args
+
+
+def main():
+    args = parse_args()
+
+    if args.result is not None and \
+            not args.result.endswith(('.pkl', '.pickle')):
+        raise ValueError('The results file must be a pkl file.')
+
+    cfg = Config.fromfile(args.config)
+    cfg.data.test.test_mode = True
+
+    # build the dataset
+    dataset = build_dataset(cfg.data.test)
+    results = mmcv.load(args.result)
+
+    if getattr(dataset, 'show', None) is not None:
+        # data loading pipeline for showing
+        eval_pipeline = cfg.get('eval_pipeline', {})
+        if eval_pipeline:
+            dataset.show(results, args.show_dir, pipeline=eval_pipeline)
+        else:
+            dataset.show(results, args.show_dir)  # use default pipeline
+    else:
+        raise NotImplementedError(
+            'Show is not implemented for dataset {}!'.format(
+                type(dataset).__name__))
+
+
+if __name__ == '__main__':
+    main()

docker-hub/BEVFormer/BEVFormer/tools/model_converters/convert_votenet_checkpoints.py

+# Copyright (c) OpenMMLab. All rights reserved.
+import argparse
+import tempfile
+import torch
+from mmcv import Config
+from mmcv.runner import load_state_dict
+
+from mmdet3d.models import build_detector
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(
+        description='MMDet3D upgrade model version(before v0.6.0) of VoteNet')
+    parser.add_argument('checkpoint', help='checkpoint file')
+    parser.add_argument('--out', help='path of the output checkpoint file')
+    args = parser.parse_args()
+    return args
+
+
+def parse_config(config_strings):
+    """Parse config from strings.
+
+    Args:
+        config_strings (string): strings of model config.
+
+    Returns:
+        Config: model config
+    """
+    temp_file = tempfile.NamedTemporaryFile()
+    config_path = f'{temp_file.name}.py'
+    with open(config_path, 'w') as f:
+        f.write(config_strings)
+
+    config = Config.fromfile(config_path)
+
+    # Update backbone config
+    if 'pool_mod' in config.model.backbone:
+        config.model.backbone.pop('pool_mod')
+
+    if 'sa_cfg' not in config.model.backbone:
+        config.model.backbone['sa_cfg'] = dict(
+            type='PointSAModule',
+            pool_mod='max',
+            use_xyz=True,
+            normalize_xyz=True)
+
+    if 'type' not in config.model.bbox_head.vote_aggregation_cfg:
+        config.model.bbox_head.vote_aggregation_cfg['type'] = 'PointSAModule'
+
+    # Update bbox_head config
+    if 'pred_layer_cfg' not in config.model.bbox_head:
+        config.model.bbox_head['pred_layer_cfg'] = dict(
+            in_channels=128, shared_conv_channels=(128, 128), bias=True)
+
+    if 'feat_channels' in config.model.bbox_head:
+        config.model.bbox_head.pop('feat_channels')
+
+    if 'vote_moudule_cfg' in config.model.bbox_head:
+        config.model.bbox_head['vote_module_cfg'] = config.model.bbox_head.pop(
+            'vote_moudule_cfg')
+
+    if config.model.bbox_head.vote_aggregation_cfg.use_xyz:
+        config.model.bbox_head.vote_aggregation_cfg.mlp_channels[0] -= 3
+
+    temp_file.close()
+
+    return config
+
+
+def main():
+    """Convert keys in checkpoints for VoteNet.
+
+    There can be some breaking changes during the development of mmdetection3d,
+    and this tool is used for upgrading checkpoints trained with old versions
+    (before v0.6.0) to the latest one.
+    """
+    args = parse_args()
+    checkpoint = torch.load(args.checkpoint)
+    cfg = parse_config(checkpoint['meta']['config'])
+    # Build the model and load checkpoint
+    model = build_detector(
+        cfg.model,
+        train_cfg=cfg.get('train_cfg'),
+        test_cfg=cfg.get('test_cfg'))
+    orig_ckpt = checkpoint['state_dict']
+    converted_ckpt = orig_ckpt.copy()
+
+    if cfg['dataset_type'] == 'ScanNetDataset':
+        NUM_CLASSES = 18
+    elif cfg['dataset_type'] == 'SUNRGBDDataset':
+        NUM_CLASSES = 10
+    else:
+        raise NotImplementedError
+
+    RENAME_PREFIX = {
+        'bbox_head.conv_pred.0': 'bbox_head.conv_pred.shared_convs.layer0',
+        'bbox_head.conv_pred.1': 'bbox_head.conv_pred.shared_convs.layer1'
+    }
+
+    DEL_KEYS = [
+        'bbox_head.conv_pred.0.bn.num_batches_tracked',
+        'bbox_head.conv_pred.1.bn.num_batches_tracked'
+    ]
+
+    EXTRACT_KEYS = {
+        'bbox_head.conv_pred.conv_cls.weight':
+        ('bbox_head.conv_pred.conv_out.weight', [(0, 2), (-NUM_CLASSES, -1)]),
+        'bbox_head.conv_pred.conv_cls.bias':
+        ('bbox_head.conv_pred.conv_out.bias', [(0, 2), (-NUM_CLASSES, -1)]),
+        'bbox_head.conv_pred.conv_reg.weight':
+        ('bbox_head.conv_pred.conv_out.weight', [(2, -NUM_CLASSES)]),
+        'bbox_head.conv_pred.conv_reg.bias':
+        ('bbox_head.conv_pred.conv_out.bias', [(2, -NUM_CLASSES)])
+    }
+
+    # Delete some useless keys
+    for key in DEL_KEYS:
+        converted_ckpt.pop(key)
+
+    # Rename keys with specific prefix
+    RENAME_KEYS = dict()
+    for old_key in converted_ckpt.keys():
+        for rename_prefix in RENAME_PREFIX.keys():
+            if rename_prefix in old_key:
+                new_key = old_key.replace(rename_prefix,
+                                          RENAME_PREFIX[rename_prefix])
+                RENAME_KEYS[new_key] = old_key
+    for new_key, old_key in RENAME_KEYS.items():
+        converted_ckpt[new_key] = converted_ckpt.pop(old_key)
+
+    # Extract weights and rename the keys
+    for new_key, (old_key, indices) in EXTRACT_KEYS.items():
+        cur_layers = orig_ckpt[old_key]
+        converted_layers = []
+        for (start, end) in indices:
+            if end != -1:
+                converted_layers.append(cur_layers[start:end])
+            else:
+                converted_layers.append(cur_layers[start:])
+        converted_layers = torch.cat(converted_layers, 0)
+        converted_ckpt[new_key] = converted_layers
+        if old_key in converted_ckpt.keys():
+            converted_ckpt.pop(old_key)
+
+    # Check the converted checkpoint by loading to the model
+    load_state_dict(model, converted_ckpt, strict=True)
+    checkpoint['state_dict'] = converted_ckpt
+    torch.save(checkpoint, args.out)
+
+
+if __name__ == '__main__':
+    main()

docker-hub/BEVFormer/BEVFormer/tools/model_converters/publish_model.py

+# Copyright (c) OpenMMLab. All rights reserved.
+import argparse
+import subprocess
+import torch
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(
+        description='Process a checkpoint to be published')
+    parser.add_argument('in_file', help='input checkpoint filename')
+    parser.add_argument('out_file', help='output checkpoint filename')
+    args = parser.parse_args()
+    return args
+
+
+def process_checkpoint(in_file, out_file):
+    checkpoint = torch.load(in_file, map_location='cpu')
+    # remove optimizer for smaller file size
+    if 'optimizer' in checkpoint:
+        del checkpoint['optimizer']
+    # if it is necessary to remove some sensitive data in checkpoint['meta'],
+    # add the code here.
+    torch.save(checkpoint, out_file)
+    sha = subprocess.check_output(['sha256sum', out_file]).decode()
+    final_file = out_file.rstrip('.pth') + '-{}.pth'.format(sha[:8])
+    subprocess.Popen(['mv', out_file, final_file])
+
+
+def main():
+    args = parse_args()
+    process_checkpoint(args.in_file, args.out_file)
+
+
+if __name__ == '__main__':
+    main()

docker-hub/BEVFormer/BEVFormer/tools/model_converters/regnet2mmdet.py

+# Copyright (c) OpenMMLab. All rights reserved.
+import argparse
+import torch
+from collections import OrderedDict
+
+
+def convert_stem(model_key, model_weight, state_dict, converted_names):
+    new_key = model_key.replace('stem.conv', 'conv1')
+    new_key = new_key.replace('stem.bn', 'bn1')
+    state_dict[new_key] = model_weight
+    converted_names.add(model_key)
+    print(f'Convert {model_key} to {new_key}')
+
+
+def convert_head(model_key, model_weight, state_dict, converted_names):
+    new_key = model_key.replace('head.fc', 'fc')
+    state_dict[new_key] = model_weight
+    converted_names.add(model_key)
+    print(f'Convert {model_key} to {new_key}')
+
+
+def convert_reslayer(model_key, model_weight, state_dict, converted_names):
+    split_keys = model_key.split('.')
+    layer, block, module = split_keys[:3]
+    block_id = int(block[1:])
+    layer_name = f'layer{int(layer[1:])}'
+    block_name = f'{block_id - 1}'
+
+    if block_id == 1 and module == 'bn':
+        new_key = f'{layer_name}.{block_name}.downsample.1.{split_keys[-1]}'
+    elif block_id == 1 and module == 'proj':
+        new_key = f'{layer_name}.{block_name}.downsample.0.{split_keys[-1]}'
+    elif module == 'f':
+        if split_keys[3] == 'a_bn':
+            module_name = 'bn1'
+        elif split_keys[3] == 'b_bn':
+            module_name = 'bn2'
+        elif split_keys[3] == 'c_bn':
+            module_name = 'bn3'
+        elif split_keys[3] == 'a':
+            module_name = 'conv1'
+        elif split_keys[3] == 'b':
+            module_name = 'conv2'
+        elif split_keys[3] == 'c':
+            module_name = 'conv3'
+        new_key = f'{layer_name}.{block_name}.{module_name}.{split_keys[-1]}'
+    else:
+        raise ValueError(f'Unsupported conversion of key {model_key}')
+    print(f'Convert {model_key} to {new_key}')
+    state_dict[new_key] = model_weight
+    converted_names.add(model_key)
+
+
+def convert(src, dst):
+    """Convert keys in pycls pretrained RegNet models to mmdet style."""
+    # load caffe model
+    regnet_model = torch.load(src)
+    blobs = regnet_model['model_state']
+    # convert to pytorch style
+    state_dict = OrderedDict()
+    converted_names = set()
+    for key, weight in blobs.items():
+        if 'stem' in key:
+            convert_stem(key, weight, state_dict, converted_names)
+        elif 'head' in key:
+            convert_head(key, weight, state_dict, converted_names)
+        elif key.startswith('s'):
+            convert_reslayer(key, weight, state_dict, converted_names)
+
+    # check if all layers are converted
+    for key in blobs:
+        if key not in converted_names:
+            print(f'not converted: {key}')
+    # save checkpoint
+    checkpoint = dict()
+    checkpoint['state_dict'] = state_dict
+    torch.save(checkpoint, dst)
+
+
+def main():
+    parser = argparse.ArgumentParser(description='Convert model keys')
+    parser.add_argument('src', help='src detectron model path')
+    parser.add_argument('dst', help='save path')
+    args = parser.parse_args()
+    convert(args.src, args.dst)
+
+
+if __name__ == '__main__':
+    main()

docker-hub/BEVFormer/BEVFormer/tools/test.py

+# ---------------------------------------------
+# Copyright (c) OpenMMLab. All rights reserved.
+# ---------------------------------------------
+#  Modified by Zhiqi Li
+# ---------------------------------------------
+import argparse
+import mmcv
+import os
+import torch
+import warnings
+from mmcv import Config, DictAction
+from mmcv.cnn import fuse_conv_bn
+from mmcv.parallel import MMDataParallel, MMDistributedDataParallel
+from mmcv.runner import (get_dist_info, init_dist, load_checkpoint,
+                         wrap_fp16_model)
+
+from mmdet3d.apis import single_gpu_test
+from mmdet3d.datasets import build_dataset
+from projects.mmdet3d_plugin.datasets.builder import build_dataloader
+from mmdet3d.models import build_model
+from mmdet.apis import set_random_seed
+from projects.mmdet3d_plugin.bevformer.apis.test import custom_multi_gpu_test
+from mmdet.datasets import replace_ImageToTensor
+import time
+import os.path as osp
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(
+        description='MMDet test (and eval) a model')
+    parser.add_argument('config', help='test config file path')
+    parser.add_argument('checkpoint', help='checkpoint file')
+    parser.add_argument('--out', help='output result file in pickle format')
+    parser.add_argument(
+        '--fuse-conv-bn',
+        action='store_true',
+        help='Whether to fuse conv and bn, this will slightly increase'
+        'the inference speed')
+    parser.add_argument(
+        '--format-only',
+        action='store_true',
+        help='Format the output results without perform evaluation. It is'
+        'useful when you want to format the result to a specific format and '
+        'submit it to the test server')
+    parser.add_argument(
+        '--eval',
+        type=str,
+        nargs='+',
+        help='evaluation metrics, which depends on the dataset, e.g., "bbox",'
+        ' "segm", "proposal" for COCO, and "mAP", "recall" for PASCAL VOC')
+    parser.add_argument('--show', action='store_true', help='show results')
+    parser.add_argument(
+        '--show-dir', help='directory where results will be saved')
+    parser.add_argument(
+        '--gpu-collect',
+        action='store_true',
+        help='whether to use gpu to collect results.')
+    parser.add_argument(
+        '--tmpdir',
+        help='tmp directory used for collecting results from multiple '
+        'workers, available when gpu-collect is not specified')
+    parser.add_argument('--seed', type=int, default=0, help='random seed')
+    parser.add_argument(
+        '--deterministic',
+        action='store_true',
+        help='whether to set deterministic options for CUDNN backend.')
+    parser.add_argument(
+        '--cfg-options',
+        nargs='+',
+        action=DictAction,
+        help='override some settings in the used config, the key-value pair '
+        'in xxx=yyy format will be merged into config file. If the value to '
+        'be overwritten is a list, it should be like key="[a,b]" or key=a,b '
+        'It also allows nested list/tuple values, e.g. key="[(a,b),(c,d)]" '
+        'Note that the quotation marks are necessary and that no white space '
+        'is allowed.')
+    parser.add_argument(
+        '--options',
+        nargs='+',
+        action=DictAction,
+        help='custom options for evaluation, the key-value pair in xxx=yyy '
+        'format will be kwargs for dataset.evaluate() function (deprecate), '
+        'change to --eval-options instead.')
+    parser.add_argument(
+        '--eval-options',
+        nargs='+',
+        action=DictAction,
+        help='custom options for evaluation, the key-value pair in xxx=yyy '
+        'format will be kwargs for dataset.evaluate() function')
+    parser.add_argument(
+        '--launcher',
+        choices=['none', 'pytorch', 'slurm', 'mpi'],
+        default='none',
+        help='job launcher')
+    parser.add_argument('--local_rank', type=int, default=0)
+    args = parser.parse_args()
+    if 'LOCAL_RANK' not in os.environ:
+        os.environ['LOCAL_RANK'] = str(args.local_rank)
+
+    if args.options and args.eval_options:
+        raise ValueError(
+            '--options and --eval-options cannot be both specified, '
+            '--options is deprecated in favor of --eval-options')
+    if args.options:
+        warnings.warn('--options is deprecated in favor of --eval-options')
+        args.eval_options = args.options
+    return args
+
+
+def main():
+    args = parse_args()
+
+    assert args.out or args.eval or args.format_only or args.show \
+        or args.show_dir, \
+        ('Please specify at least one operation (save/eval/format/show the '
+         'results / save the results) with the argument "--out", "--eval"'
+         ', "--format-only", "--show" or "--show-dir"')
+
+    if args.eval and args.format_only:
+        raise ValueError('--eval and --format_only cannot be both specified')
+
+    if args.out is not None and not args.out.endswith(('.pkl', '.pickle')):
+        raise ValueError('The output file must be a pkl file.')
+
+    cfg = Config.fromfile(args.config)
+    if args.cfg_options is not None:
+        cfg.merge_from_dict(args.cfg_options)
+    # import modules from string list.
+    if cfg.get('custom_imports', None):
+        from mmcv.utils import import_modules_from_strings
+        import_modules_from_strings(**cfg['custom_imports'])
+
+    # import modules from plguin/xx, registry will be updated
+    if hasattr(cfg, 'plugin'):
+        if cfg.plugin:
+            import importlib
+            if hasattr(cfg, 'plugin_dir'):
+                plugin_dir = cfg.plugin_dir
+                _module_dir = os.path.dirname(plugin_dir)
+                _module_dir = _module_dir.split('/')
+                _module_path = _module_dir[0]
+
+                for m in _module_dir[1:]:
+                    _module_path = _module_path + '.' + m
+                print(_module_path)
+                plg_lib = importlib.import_module(_module_path)
+            else:
+                # import dir is the dirpath for the config file
+                _module_dir = os.path.dirname(args.config)
+                _module_dir = _module_dir.split('/')
+                _module_path = _module_dir[0]
+                for m in _module_dir[1:]:
+                    _module_path = _module_path + '.' + m
+                print(_module_path)
+                plg_lib = importlib.import_module(_module_path)
+
+    # set cudnn_benchmark
+    if cfg.get('cudnn_benchmark', False):
+        torch.backends.cudnn.benchmark = True
+    # set tf32
+    if cfg.get('close_tf32', False):
+        torch.backends.cuda.matmul.allow_tf32 = False
+        torch.backends.cudnn.allow_tf32 = False
+
+    cfg.model.pretrained = None
+    # in case the test dataset is concatenated
+    samples_per_gpu = 1
+    if isinstance(cfg.data.test, dict):
+        cfg.data.test.test_mode = True
+        samples_per_gpu = cfg.data.test.pop('samples_per_gpu', 1)
+        if samples_per_gpu > 1:
+            # Replace 'ImageToTensor' to 'DefaultFormatBundle'
+            cfg.data.test.pipeline = replace_ImageToTensor(
+                cfg.data.test.pipeline)
+    elif isinstance(cfg.data.test, list):
+        for ds_cfg in cfg.data.test:
+            ds_cfg.test_mode = True
+        samples_per_gpu = max(
+            [ds_cfg.pop('samples_per_gpu', 1) for ds_cfg in cfg.data.test])
+        if samples_per_gpu > 1:
+            for ds_cfg in cfg.data.test:
+                ds_cfg.pipeline = replace_ImageToTensor(ds_cfg.pipeline)
+
+    # 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)
+
+    # set random seeds
+    if args.seed is not None:
+        set_random_seed(args.seed, deterministic=args.deterministic)
+
+    # build the dataloader
+    dataset = build_dataset(cfg.data.test)
+    data_loader = build_dataloader(
+        dataset,
+        samples_per_gpu=samples_per_gpu,
+        workers_per_gpu=cfg.data.workers_per_gpu,
+        dist=distributed,
+        shuffle=False,
+        nonshuffler_sampler=cfg.data.nonshuffler_sampler,
+    )
+
+    # build the model and load checkpoint
+    cfg.model.train_cfg = None
+    model = build_model(cfg.model, test_cfg=cfg.get('test_cfg'))
+    fp16_cfg = cfg.get('fp16', None)
+    if fp16_cfg is not None:
+        wrap_fp16_model(model)
+    checkpoint = load_checkpoint(model, args.checkpoint, map_location='cpu')
+    if args.fuse_conv_bn:
+        model = fuse_conv_bn(model)
+    # old versions did not save class info in checkpoints, this walkaround is
+    # for backward compatibility
+    if 'CLASSES' in checkpoint.get('meta', {}):
+        model.CLASSES = checkpoint['meta']['CLASSES']
+    else:
+        model.CLASSES = dataset.CLASSES
+    # palette for visualization in segmentation tasks
+    if 'PALETTE' in checkpoint.get('meta', {}):
+        model.PALETTE = checkpoint['meta']['PALETTE']
+    elif hasattr(dataset, 'PALETTE'):
+        # segmentation dataset has `PALETTE` attribute
+        model.PALETTE = dataset.PALETTE
+
+    if not distributed:
+        assert False
+        # model = MMDataParallel(model, device_ids=[0])
+        # outputs = single_gpu_test(model, data_loader, args.show, args.show_dir)
+    else:
+        model = MMDistributedDataParallel(
+            model.cuda(),
+            device_ids=[torch.cuda.current_device()],
+            broadcast_buffers=False)
+        outputs = custom_multi_gpu_test(model, data_loader, args.tmpdir,
+                                        args.gpu_collect)
+
+    rank, _ = get_dist_info()
+    if rank == 0:
+        if args.out:
+            print(f'\nwriting results to {args.out}')
+            assert False
+            #mmcv.dump(outputs['bbox_results'], args.out)
+        kwargs = {} if args.eval_options is None else args.eval_options
+        kwargs['jsonfile_prefix'] = osp.join('test', args.config.split(
+            '/')[-1].split('.')[-2], time.ctime().replace(' ', '_').replace(':', '_'))
+        if args.format_only:
+            dataset.format_results(outputs, **kwargs)
+
+        if args.eval:
+            eval_kwargs = cfg.get('evaluation', {}).copy()
+            # hard-code way to remove EvalHook args
+            for key in [
+                    'interval', 'tmpdir', 'start', 'gpu_collect', 'save_best',
+                    'rule'
+            ]:
+                eval_kwargs.pop(key, None)
+            eval_kwargs.update(dict(metric=args.eval, **kwargs))
+
+            print(dataset.evaluate(outputs, **eval_kwargs))
+
+
+if __name__ == '__main__':
+    torch.multiprocessing.set_start_method('fork')
+    main()

docker-hub/BEVFormer/BEVFormer/tools/train.py

+# ---------------------------------------------
+# Copyright (c) OpenMMLab. All rights reserved.
+# ---------------------------------------------
+#  Modified by Zhiqi Li
+# ---------------------------------------------
+ 
+from __future__ import division
+
+import argparse
+import copy
+import mmcv
+import os
+import time
+import torch
+import warnings
+from mmcv import Config, DictAction
+from mmcv.runner import get_dist_info, init_dist
+from os import path as osp
+
+from mmdet import __version__ as mmdet_version
+from mmdet3d import __version__ as mmdet3d_version
+#from mmdet3d.apis import train_model
+
+from mmdet3d.datasets import build_dataset
+from mmdet3d.models import build_model
+from mmdet3d.utils import collect_env, get_root_logger
+from mmdet.apis import set_random_seed
+from mmseg import __version__ as mmseg_version
+
+from mmcv.utils import TORCH_VERSION, digit_version
+
+
+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(
+        '--no-validate',
+        action='store_true',
+        help='whether not to evaluate the checkpoint during training')
+    group_gpus = parser.add_mutually_exclusive_group()
+    group_gpus.add_argument(
+        '--gpus',
+        type=int,
+        help='number of gpus to use '
+        '(only applicable to non-distributed training)')
+    group_gpus.add_argument(
+        '--gpu-ids',
+        type=int,
+        nargs='+',
+        help='ids of gpus to use '
+        '(only applicable to non-distributed training)')
+    parser.add_argument('--seed', type=int, default=0, help='random seed')
+    parser.add_argument(
+        '--deterministic',
+        action='store_true',
+        help='whether to set deterministic options for CUDNN backend.')
+    parser.add_argument(
+        '--options',
+        nargs='+',
+        action=DictAction,
+        help='override some settings in the used config, the key-value pair '
+        'in xxx=yyy format will be merged into config file (deprecate), '
+        'change to --cfg-options instead.')
+    parser.add_argument(
+        '--cfg-options',
+        nargs='+',
+        action=DictAction,
+        help='override some settings in the used config, the key-value pair '
+        'in xxx=yyy format will be merged into config file. If the value to '
+        'be overwritten is a list, it should be like key="[a,b]" or key=a,b '
+        'It also allows nested list/tuple values, e.g. key="[(a,b),(c,d)]" '
+        'Note that the quotation marks are necessary and that no white space '
+        'is allowed.')
+    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)
+
+    if args.options and args.cfg_options:
+        raise ValueError(
+            '--options and --cfg-options cannot be both specified, '
+            '--options is deprecated in favor of --cfg-options')
+    if args.options:
+        warnings.warn('--options is deprecated in favor of --cfg-options')
+        args.cfg_options = args.options
+
+    return args
+
+
+def main():
+    args = parse_args()
+
+    cfg = Config.fromfile(args.config)
+    if args.cfg_options is not None:
+        cfg.merge_from_dict(args.cfg_options)
+    # import modules from string list.
+    if cfg.get('custom_imports', None):
+        from mmcv.utils import import_modules_from_strings
+        import_modules_from_strings(**cfg['custom_imports'])
+
+    # import modules from plguin/xx, registry will be updated
+    if hasattr(cfg, 'plugin'):
+        if cfg.plugin:
+            import importlib
+            if hasattr(cfg, 'plugin_dir'):
+                plugin_dir = cfg.plugin_dir
+                _module_dir = os.path.dirname(plugin_dir)
+                _module_dir = _module_dir.split('/')
+                _module_path = _module_dir[0]
+
+                for m in _module_dir[1:]:
+                    _module_path = _module_path + '.' + m
+                print(_module_path)
+                plg_lib = importlib.import_module(_module_path)
+            else:
+                # import dir is the dirpath for the config file
+                _module_dir = os.path.dirname(args.config)
+                _module_dir = _module_dir.split('/')
+                _module_path = _module_dir[0]
+                for m in _module_dir[1:]:
+                    _module_path = _module_path + '.' + m
+                print(_module_path)
+                plg_lib = importlib.import_module(_module_path)
+
+            from projects.mmdet3d_plugin.bevformer.apis.train import custom_train_model
+    # set cudnn_benchmark
+    if cfg.get('cudnn_benchmark', False):
+        torch.backends.cudnn.benchmark = True
+    # set tf32
+    if cfg.get('close_tf32', False):
+        torch.backends.cuda.matmul.allow_tf32 = False
+        torch.backends.cudnn.allow_tf32 = False
+
+    # work_dir is determined in this priority: CLI > segment in file > filename
+    if args.work_dir is not None:
+        # update configs according to CLI args if args.work_dir is not None
+        cfg.work_dir = args.work_dir
+    elif cfg.get('work_dir', None) is None:
+        # use config filename as default work_dir if cfg.work_dir is None
+        cfg.work_dir = osp.join('./work_dirs',
+                                osp.splitext(osp.basename(args.config))[0])
+    # if args.resume_from is not None:
+    if args.resume_from is not None and osp.isfile(args.resume_from):
+        cfg.resume_from = args.resume_from
+    if args.gpu_ids is not None:
+        cfg.gpu_ids = args.gpu_ids
+    else:
+        cfg.gpu_ids = range(1) if args.gpus is None else range(args.gpus)
+    if digit_version(TORCH_VERSION) == digit_version('1.8.1') and cfg.optimizer['type'] == 'AdamW':
+        cfg.optimizer['type'] = 'AdamW2' # fix bug in Adamw
+    if args.autoscale_lr:
+        # apply the linear scaling rule (https://arxiv.org/abs/1706.02677)
+        cfg.optimizer['lr'] = cfg.optimizer['lr'] * len(cfg.gpu_ids) / 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)
+        # re-set gpu_ids with distributed training mode
+        _, world_size = get_dist_info()
+        cfg.gpu_ids = range(world_size)
+
+    # create work_dir
+    mmcv.mkdir_or_exist(osp.abspath(cfg.work_dir))
+    # dump config
+    cfg.dump(osp.join(cfg.work_dir, osp.basename(args.config)))
+    # init the logger before other steps
+    timestamp = time.strftime('%Y%m%d_%H%M%S', time.localtime())
+    log_file = osp.join(cfg.work_dir, f'{timestamp}.log')
+    # specify logger name, if we still use 'mmdet', the output info will be
+    # filtered and won't be saved in the log_file
+    # TODO: ugly workaround to judge whether we are training det or seg model
+    if cfg.model.type in ['EncoderDecoder3D']:
+        logger_name = 'mmseg'
+    else:
+        logger_name = 'mmdet'
+    logger = get_root_logger(
+        log_file=log_file, log_level=cfg.log_level, name=logger_name)
+
+    # 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([(f'{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
+    meta['config'] = cfg.pretty_text
+
+    # log some basic info
+    logger.info(f'Distributed training: {distributed}')
+    logger.info(f'Config:\n{cfg.pretty_text}')
+
+    # set random seeds
+    if args.seed is not None:
+        logger.info(f'Set random seed to {args.seed}, '
+                    f'deterministic: {args.deterministic}')
+        set_random_seed(args.seed, deterministic=args.deterministic)
+    cfg.seed = args.seed
+    meta['seed'] = args.seed
+    meta['exp_name'] = osp.basename(args.config)
+
+    model = build_model(
+        cfg.model,
+        train_cfg=cfg.get('train_cfg'),
+        test_cfg=cfg.get('test_cfg'))
+    model.init_weights()
+
+    logger.info(f'Model:\n{model}')
+    datasets = [build_dataset(cfg.data.train)]
+    if len(cfg.workflow) == 2:
+        val_dataset = copy.deepcopy(cfg.data.val)
+        # in case we use a dataset wrapper
+        if 'dataset' in cfg.data.train:
+            val_dataset.pipeline = cfg.data.train.dataset.pipeline
+        else:
+            val_dataset.pipeline = cfg.data.train.pipeline
+        # set test_mode=False here in deep copied config
+        # which do not affect AP/AR calculation later
+        # refer to https://mmdetection3d.readthedocs.io/en/latest/tutorials/customize_runtime.html#customize-workflow  # noqa
+        val_dataset.test_mode = False
+        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=mmdet_version,
+            mmseg_version=mmseg_version,
+            mmdet3d_version=mmdet3d_version,
+            config=cfg.pretty_text,
+            CLASSES=datasets[0].CLASSES,
+            PALETTE=datasets[0].PALETTE  # for segmentors
+            if hasattr(datasets[0], 'PALETTE') else None)
+    # add an attribute for visualization convenience
+    model.CLASSES = datasets[0].CLASSES
+
+    torch.backends.cudnn.benchmark = True  # 启用自动寻找最优卷积算法
+    torch.backends.cudnn.deterministic = False  # 允许非确定性算法提升速度
+
+    custom_train_model(
+        model,
+        datasets,
+        cfg,
+        distributed=distributed,
+        validate=(not args.no_validate),
+        timestamp=timestamp,
+        meta=meta)
+
+
+if __name__ == '__main__':
+    torch.multiprocessing.set_start_method('fork')
+    main()