nuscenes数据集

数据格式：

1.时间戳取值为秒*1000000，nuscenes devkit在解析时以精度作为转换。

2.单位旋转矩阵对应的四元数为：[1,0,0,0]

3.传感器外参为传感器到ego的转换

验证数据格式：

1.检测结果格式：

https://www.nuscenes.org/object-detection/?externalData=all&mapData=all&modalities=Anyhttps://www.nuscenes.org/object-detection/?externalData=all&mapData=all&modalities=Any

2.跟踪结果格式：

跟踪结果格式：https://www.nuscenes.org/tracking/?externalData=all&mapData=all&modalities=Anyhttps://www.nuscenes.org/tracking/?externalData=all&mapData=all&modalities=Any

submission {
    "meta": {
        "use_camera":   <bool>  -- Whether this submission uses camera data as an input.
        "use_lidar":    <bool>  -- Whether this submission uses lidar data as an input.
        "use_radar":    <bool>  -- Whether this submission uses radar data as an input.
        "use_map":      <bool>  -- Whether this submission uses map data as an input.
        "use_external": <bool>  -- Whether this submission uses external data as an input.
    },
    "results": {
        sample_token <str>: List[sample_result] -- Maps each sample_token to a list of sample_results.
    }
}

sample_result {
    "sample_token":   <str>         -- Foreign key. Identifies the sample/keyframe for which objects are detected.
    "translation":    <float> [3]   -- Estimated bounding box location in meters in the global frame: center_x, center_y, center_z.
    "size":           <float> [3]   -- Estimated bounding box size in meters: width, length, height.
    "rotation":       <float> [4]   -- Estimated bounding box orientation as quaternion in the global frame: w, x, y, z.
    "velocity":       <float> [2]   -- Estimated bounding box velocity in m/s in the global frame: vx, vy.
    "tracking_id":    <str>         -- Unique object id that is used to identify an object track across samples.
    "tracking_name":  <str>         -- The predicted class for this sample_result, e.g. car, pedestrian.
                                       Note that the tracking_name cannot change throughout a track.
    "tracking_score": <float>       -- Object prediction score between 0 and 1 for the class identified by tracking_name.
                                       We average over frame level scores to compute the track level score.
                                       The score is used to determine positive and negative tracks via thresholding.
}

devkit：

验证方法：

cfg_ = config_factory('tracking_nips_2019')
nusc_eval = TrackingEval(config=cfg_,result_path=elv_json,eval_set='train',output_dir="/data/nuscenes/eval",nusc_version='V1.0-trainval',nusc_dataroot='/data/pipeline_data/liujiao/bicv/nuscenes/data0202',)
nusc_eval.main()

可视化：

nusc = NuScenes(version='V1.0-trainval', dataroot='/data/pipeline_data/liujiao/bicv/nuscenes/data0202', verbose=True)
nusc.list_scenes()
my_scene = nusc.scene[0]
first_sample_token = my_scene['first_sample_token']  #获取第一个sample的token值
my_sample = nusc.get("sample",first_sample_token)
aa = nusc.list_sample(my_sample["token"])
annotation_token = my_sample["anns"][0]
annotation_metadata = nusc.get("sample_annotation",annotation_token)
nusc.render_annotation(annotation_token,out_path="annotation.jpg")
nusc.render_sample_data(my_sample["data"]["CAM_FRONT_LEFT"],out_path="camfrontleft.jpg")
my_instace = nusc.instance[0]
nusc.render_annotation(my_instace['first_annotation_token'],out_path="first_annotation.jpg")

二进制pcd文件

nuscenes数据集中激光雷达点云文件为二进制文件，pcd转二进制文件方法：

def read_pcd_binary(pcd_file,bin_save_file):
    with open(pcd_file, 'rb') as f:
        # 读取pcd文件头信息
        for line in f:
            line = line.decode('utf-8')
            if line.startswith("DATA"):
                break
        # 从文件内容中读取三维坐标点云数据
        dtype = np.dtype(
            [('x', np.float32), ('y', np.float32), ('z', np.float32), ('i', np.uint16), ('r', np.uint16),
                ('t', np.float64)])
        data = np.fromfile(f, dtype=dtype)
        data = np.array(data.tolist())
        data = np.delete(data, np.where(np.isnan(data))[0], axis=0)
        data = np.delete(data, 5, axis=1)
        binary_data = data.astype(np.float32).tobytes()
        with open(bin_save_file, 'wb') as f:
            f.write(binary_data)

点云二进制文件可视化代码：

import os
import open3d as o3d
import numpy as np

def read_bin_point_cloud(bin_file):
    try:
        with open(bin_file, 'rb') as f:
            data = np.fromfile(f, dtype=np.float32)
    except FileNotFoundError:
        print(f"Error: File '{bin_file}' not found.")
        return None
    points = data.reshape(-1, 5)
    points = points[:,:3]

    point_cloud = o3d.geometry.PointCloud()
    point_cloud.points = o3d.utility.Vector3dVector(points)

    return point_cloud

def visualize_point_cloud(point_cloud):
    vis = o3d.visualization.Visualizer()
    vis.create_window()
    vis.add_geometry(point_cloud)
    render_options = vis.get_render_option()
    render_options.point_size = 2
    vis.run()
    vis.destroy_window()

if __name__ == "__main__":
    bin_file_path = "output02.pcd.bin"
    point_cloud = read_bin_point_cloud(bin_file_path)
    if point_cloud:
        visualize_point_cloud(point_cloud)