自动驾驶云服务 OCTOPUS-预审核模型文件:预审核结果格式说明
预审核结果格式说明
审核完毕后,需要按照规定格式组织预标注结果,并保存在特定路径(TARGET_RESULT_DIR)下的json文件中。路径要求见镜像制作(标注)。
Json文件内容组织结构如下所示,labels字段中保存每个对象的标注信息、审核模型预测信息(predict_infos)和审核结果信息(inspection)。
{ "labels":[{ #1. 此对象的标注信息 (直接从源数据labels.json中获取),如果未标注出此对象,则无此部分信息 … … #2.此对象的模型预测信息 ,如果模型未预测出此对象,则无此部分信息 "predict_infos": { #形状坐标信息 #对象类别名称 #额外属性信息 } #3.审核结果,如果未审核此对象,则无此部分信息 "inspection": { #字段名称取自OCTPS_INSPECTION_ATTRI_DIR文件 } }, … … ] }
其中3D大规模点云分割任务还包含“labels_ext”和“predict_labels_ext”字段,具体参考“3D大规模点云分割”。
{ "labels":[], "labels_ext":{ } "predict_labels":[] }
以2D目标检测为例,完整json结果文件样例如下:
{ "labels": [ { #1. 此对象的标注信息(直接从源数据labels.json中获取) "label_meta_id": 1846, "bndbox": { "ymin": 545.4334, "xmin": 1158.3188, "ymax": 705.71844, "xmax": 1436.3274 }, "name": "框0504", "shape_type": "bndbox", "serial_number": 2, "label_object_id": 2, "attribute": "{\"优先级\":\"1\"}", "label_meta_name": "框0504", #2.此对象的模型预测信息 "predict_infos": { "bndbox": { "ymin": 545.4334, "xmin": 1158.3188, "ymax": 725.71844, "xmax": 1456.3274 }, "label_meta_name": "框0504", "attribute": "{\"优先级\":\"1\"}" }, #3.审核结果 "inspection": { #字段名称取自OCTPS_INSPECTION_ATTRI_DIR文件 "miss_label_error": false, "vehicle_direction_error": false, "error_desc": "无效", "attribute_error": true, "out_range_label_error": true, "anchor_error": false, "classification_error": false, "extra_label_error": false } } ] }
不同类型的标注对象形状基本信息所需格式不同。下面为各类标注对象predict_infos的字段说明:
2D目标检测:
{"predict_infos": { "bndbox": { "ymin": 545.4334, "xmin": 1158.3188, "ymax": 725.71844, "xmax": 1456.3274 }, "label_meta_name": "框0504", "attribute": "{\"优先级\":\"1\"}" } }
2D语义分割:
{"predict_infos": { "polygon": { "size": 3, "points": [ { "xpoint": 135.03, "ypoint": 482.94937 }, { "xpoint": 84.318344, "ypoint": 554.4891 }, { "xpoint": 135.03, "ypoint": 482.94937 } ] }, "label_meta_name": "多边形0504", "attribute": "{\"优先级\":\"1\"}" } }
2D车道线:
{"predict_infos": { "line": { "size": 3, "points": [ { "xpoint": 901.138, "ypoint": 553.583 }, { "xpoint": 741.36, "ypoint": 630.367 }, { "xpoint": 618.153, "ypoint": 681.566 } ] }, "label_meta_name": "线0504", "attribute": "{\"优先级\":\"1\"}" } }
3D目标检测:
{ "predict_infos": { "label_meta_name": "Car", "cube_3d": { "rotation": { "x": 0.0, "y": 0.0, "z": 0.08726646 }, "location": { "x": -40.23651584555386, "y": 1.2362389665094042, "z": -0.8413386615781039 }, "attribute": "{}", "dimensions": { "length": 4.459540762142082, "width": 1.4870339632034302, "height": 1.4895729290943762 } } } }
3D大规模点云分割:
{"predict_infos": { "polygon_3d_v2": { "ascii_char": "2" }, "name": "car", } }
3D大规模点云分割完整样例
{ "labels": [ { "label_meta_id": 4867, "create_time": 0, "polygon_3d_v2": { "ascii_char": "3" }, "name": "car", "shape_type": "polygon_3d_v2", "serial_number": 0, "label_object_id": -1, "attribute": "", "label_meta_name": "car", "inspection": { "miss_label_error": false, "vehicle_direction_error": false, "error_desc": "", "attribute_error": false, "out_range_label_error": false, "anchor_error": false, "classification_error": false, "extra_label_error": false }, "predict_infos": { "polygon_3d_v2": { "ascii_char": "2" }, "name": "car" } }, { "predict_infos": { "polygon_3d_v2": { "ascii_char": "4" }, "name": "van" } } ], "labels_ext": { "ascii_string": "3333333333 3333333333" }, "predict_labels_ext": { "ascii_string": "222222222244444 2222222222" } }
labels_ext中保存点云中每个点的标注类别,具体内容说明参考OCTOPUS数据集格式说明。predict_labels_ext中保存点云中每个点的模型预测类别。
3D语义分割审核结果可视化说明:针对有审核属性错误的标注对象,展示该标注对象对应位置点的预测类别。