参考
关于coco的格式
https://detectron2.readthedocs.io/en/latest/tutorials/datasets.html#register-a-dataset
注册并训练自己的数据集合/qq_29750461/article/details/106761382
/developer/article/1960793
coco api /qq_41709370/article/details/108471072
注意这里如果出现说自己的数据没有注册的错误，需要在main里面加上函数plain_register_dataset()注册自己的数据集
数据转换成coco格式
一些坑：
bbox对于coco格式来说[x,y,width,height] x,y是左上角的坐标
file_name是图片的绝对路径 要写成xxx.jpg
有些coco文件里说不是必须的字段但是如果没有会报keyerror
anntations里比如area、iscrowd 可以设置默认值 area可以是bbox的w*h
读取图片shape[0]是height shape[1]是witdh
height = img.shape[0]width = img.shape[1]
注册数据集
#!/usr/bin/env python# Copyright (c) Facebook, Inc. and its affiliates."""A main training script.This scripts reads a given config file and runs the training or evaluation.It is an entry point that is made to train standard models in detectron2.In order to let one script support training of many models,this script contains logic that are specific to these built-in models and thereforemay not be suitable for your own project.For example, your research project perhaps only needs a single "evaluator".Therefore, we recommend you to use detectron2 as an library and takethis file as an example of how to use the library.You may want to write your own script with your datasets and other customizations."""import loggingimport osfrom collections import OrderedDictimport cv2from detectron2.utils.visualizer import Visualizerimport m as commfrom detectron2.checkpoint import DetectionCheckpointerfrom detectron2.config import get_cfgfrom detectron2.data import MetadataCatalogfrom detectron2.engine import DefaultTrainer, default_argument_parser, default_setup, hooks, launchfrom detectron2.evaluation import (CityscapesInstanceEvaluator,CityscapesSemSegEvaluator,COCOEvaluator,COCOPanopticEvaluator,DatasetEvaluators,LVISEvaluator,PascalVOCDetectionEvaluator,SemSegEvaluator,verify_results,)from detectron2.modeling import GeneralizedRCNNWithTTAfrom detectron2.data import DatasetCatalog, MetadataCatalogfrom detectron2.data.datasets.coco import load_coco_jsonimport pycocotools#声明类别，尽量保持CLASS_NAMES =["background", "military"]# 数据集路径DATASET_ROOT = '/home/szr/new_dete/detectron2/datasets'ANN_ROOT = os.path.join(DATASET_ROOT, 'COCOformat')TRAIN_PATH = os.path.join(DATASET_ROOT, 'JPEGImages')VAL_PATH = os.path.join(DATASET_ROOT, 'JPEGImages')TRAIN_JSON = os.path.join(ANN_ROOT, 'train.json')VAL_JSON = os.path.join(ANN_ROOT, 'val.json')# VAL_JSON = os.path.join(ANN_ROOT, 'test.json')# 声明数据集的子集PREDEFINED_SPLITS_DATASET = {"coco_my_train": (TRAIN_PATH, TRAIN_JSON),"coco_my_val": (VAL_PATH, VAL_JSON),}#===========以下有两种注册数据集的方法，本人直接用的第二个plain_register_dataset的方式 也可以用register_dataset的形式==================#注册数据集（这一步就是将自定义数据集注册进Detectron2）# 注册数据集和元数据def plain_register_dataset():#训练集DatasetCatalog.register("coco_my_train", lambda: load_coco_json(TRAIN_JSON, TRAIN_PATH))MetadataCatalog.get("coco_my_train").set(thing_classes=CLASS_NAMES, # 可以选择开启，但是不能显示中文，这里需要注意，中文的话最好关闭evaluator_type='coco', # 指定评估方式json_file=TRAIN_JSON,image_root=TRAIN_PATH)#DatasetCatalog.register("coco_my_val", lambda: load_coco_json(VAL_JSON, VAL_PATH, "coco__val"))#验证/测试集DatasetCatalog.register("coco_my_val", lambda: load_coco_json(VAL_JSON, VAL_PATH))MetadataCatalog.get("coco_my_val").set(thing_classes=CLASS_NAMES, # 可以选择开启，但是不能显示中文，这里需要注意，中文的话最好关闭evaluator_type='coco', # 指定评估方式json_file=VAL_JSON,image_root=VAL_PATH)# 查看数据集标注，可视化检查数据集标注是否正确，#这个也可以自己写脚本判断，其实就是判断标注框是否超越图像边界#可选择使用此方法def checkout_dataset_annotation(name="coco_my_val"):#dataset_dicts = load_coco_json(TRAIN_JSON, TRAIN_PATH, name)dataset_dicts = load_coco_json(TRAIN_JSON, TRAIN_PATH)print(len(dataset_dicts))for i, d in enumerate(dataset_dicts,0):#print(d)img = cv2.imread(d["file_name"])visualizer = Visualizer(img[:, :, ::-1], metadata=MetadataCatalog.get(name), scale=1.5)vis = visualizer.draw_dataset_dict(d)#cv2.imshow('show', vis.get_image()[:, :, ::-1])cv2.imwrite('out/'+str(i) + '.jpg',vis.get_image()[:, :, ::-1])#cv2.waitKey(0)if i == 200:breakdef build_evaluator(cfg, dataset_name, output_folder=None):"""Create evaluator(s) for a given dataset.This uses the special metadata "evaluator_type" associated with each builtin dataset.For your own dataset, you can simply create an evaluator manually in yourscript and do not have to worry about the hacky if-else logic here."""if output_folder is None:output_folder = os.path.join(cfg.OUTPUT_DIR, "inference")evaluator_list = []evaluator_type = MetadataCatalog.get(dataset_name).evaluator_typeif evaluator_type in ["sem_seg", "coco_panoptic_seg"]:evaluator_list.append(SemSegEvaluator(dataset_name,distributed=True,output_dir=output_folder,))if evaluator_type in ["coco", "coco_panoptic_seg"]:evaluator_list.append(COCOEvaluator(dataset_name, output_dir=output_folder))if evaluator_type == "coco_panoptic_seg":evaluator_list.append(COCOPanopticEvaluator(dataset_name, output_folder))if evaluator_type == "cityscapes_instance":return CityscapesInstanceEvaluator(dataset_name)if evaluator_type == "cityscapes_sem_seg":return CityscapesSemSegEvaluator(dataset_name)elif evaluator_type == "pascal_voc":return PascalVOCDetectionEvaluator(dataset_name)elif evaluator_type == "lvis":return LVISEvaluator(dataset_name, output_dir=output_folder)if len(evaluator_list) == 0:raise NotImplementedError("no Evaluator for the dataset {} with the type {}".format(dataset_name, evaluator_type))elif len(evaluator_list) == 1:return evaluator_list[0]return DatasetEvaluators(evaluator_list)class Trainer(DefaultTrainer):"""We use the "DefaultTrainer" which contains pre-defined default logic forstandard training workflow. They may not work for you, especially if youare working on a new research project. In that case you can write yourown training loop. You can use "tools/plain_train_net.py" as an example."""@classmethoddef build_evaluator(cls, cfg, dataset_name, output_folder=None):return build_evaluator(cfg, dataset_name, output_folder)@classmethoddef test_with_TTA(cls, cfg, model):logger = logging.getLogger("detectron2.trainer")# In the end of training, run an evaluation with TTA# Only support some R-CNN models.logger.info("Running inference with test-time augmentation ...")model = GeneralizedRCNNWithTTA(cfg, model)evaluators = [cls.build_evaluator(cfg, name, output_folder=os.path.join(cfg.OUTPUT_DIR, "inference_TTA"))for name in cfg.DATASETS.TEST]res = cls.test(cfg, model, evaluators)res = OrderedDict({k + "_TTA": v for k, v in res.items()})return res# def setup(args):#"""#Create configs and perform basic setups.#"""## cfg = get_cfg()## cfg.merge_from_file(args.config_file)## cfg.merge_from_list(args.opts)## cfg.freeze()## default_setup(cfg, args)## return cfgdef setup(args):"""Create configs and perform basic setups."""cfg = get_cfg()args.config_file = "/home/szr/new_dete/detectron2/configs/COCO-Detection/retinanet_R_50_FPN_3x.yaml"cfg.merge_from_file(args.config_file) # 从config file 覆盖配置cfg.merge_from_list(args.opts) # 从CLI参数 覆盖配置# 更改配置参数cfg.DATASETS.TRAIN = ("coco_my_train",) # 训练数据集名称cfg.DATASETS.TEST = ("coco_my_val",)cfg.DATALOADER.NUM_WORKERS = 4 # 单线程cfg.INPUT.CROP.ENABLED = Truecfg.INPUT.MAX_SIZE_TRAIN = 640 # 训练图片输入的最大尺寸cfg.INPUT.MAX_SIZE_TEST = 640 # 测试数据输入的最大尺寸cfg.INPUT.MIN_SIZE_TRAIN = (512, 768) # 训练图片输入的最小尺寸，可以设定为多尺度训练cfg.INPUT.MIN_SIZE_TEST = 640# cfg.INPUT.MIN_SIZE_TRAIN_SAMPLING，其存在两种配置，分别为 choice 与 range ：# range 让图像的短边从 512-768随机选择# choice ： 把输入图像转化为指定的，有限的几种图片大小进行训练，即短边只能为 512或者768cfg.INPUT.MIN_SIZE_TRAIN_SAMPLING = 'range'# 本句一定要看下注释！！！！！！！！cfg.MODEL.RETINANET.NUM_CLASSES = 2 # 类别数+1（因为有background，也就是你的 cate id 从 1 开始，如果您的数据集Json下标从 0 开始，这个改为您对应的类别就行，不用再加背景类！！！！！）# cfg.MODEL.WEIGHTS="/home/yourstorePath/.pth"#这个地址可以自己下载或者它会自己联网下载cfg.MODEL.WEIGHTS = "detectron2://COCO-Detection/retinanet_R_50_FPN_3x/190397829/model_final_5bd44e.pkl" # 预训练模型权重cfg.SOLVER.IMS_PER_BATCH = 4 # batch_size=2; iters_in_one_epoch = dataset_imgs/batch_size# 根据训练数据总数目以及batch_size，计算出每个epoch需要的迭代次数# 9000为你的训练数据的总数目，可自定义ITERS_IN_ONE_EPOCH = int(1120 / cfg.SOLVER.IMS_PER_BATCH)# 指定最大迭代次数cfg.SOLVER.MAX_ITER = (ITERS_IN_ONE_EPOCH * 12) - 1 # 12 epochs，# 初始学习率cfg.SOLVER.BASE_LR = 0.002# 优化器动能cfg.SOLVER.MOMENTUM = 0.9# 权重衰减cfg.SOLVER.WEIGHT_DECAY = 0.0001cfg.SOLVER.WEIGHT_DECAY_NORM = 0.0# 学习率衰减倍数cfg.SOLVER.GAMMA = 0.1# 迭代到指定次数，学习率进行衰减cfg.SOLVER.STEPS = (7000,)# 在训练之前，会做一个热身运动，学习率慢慢增加初始学习率cfg.SOLVER.WARMUP_FACTOR = 1.0 / 1000# 热身迭代次数cfg.SOLVER.WARMUP_ITERS = 1000cfg.SOLVER.WARMUP_METHOD = "linear"# 保存模型文件的命名数据减1cfg.SOLVER.CHECKPOINT_PERIOD = ITERS_IN_ONE_EPOCH - 1# 迭代到指定次数，进行一次评估cfg.TEST.EVAL_PERIOD = ITERS_IN_ONE_EPOCH# cfg.TEST.EVAL_PERIOD = 100# cfg.merge_from_file(args.config_file)# cfg.merge_from_list(args.opts)cfg.freeze()default_setup(cfg, args)return cfgdef main(args):cfg = setup(args)if args.eval_only:model = Trainer.build_model(cfg)DetectionCheckpointer(model, save_dir=cfg.OUTPUT_DIR).resume_or_load(cfg.MODEL.WEIGHTS, resume=args.resume)res = Trainer.test(cfg, model)if cfg.TEST.AUG.ENABLED:res.update(Trainer.test_with_TTA(cfg, model))if comm.is_main_process():verify_results(cfg, res)return res"""If you'd like to do anything fancier than the standard training logic,consider writing your own training loop (see plain_train_net.py) orsubclassing the trainer."""trainer = Trainer(cfg)trainer.resume_or_load(resume=args.resume)if cfg.TEST.AUG.ENABLED:trainer.register_hooks([hooks.EvalHook(0, lambda: trainer.test_with_TTA(cfg, trainer.model))])return trainer.train()if __name__ == "__main__":args = default_argument_parser().parse_args()plain_register_dataset()print("Command Line Args:", args)launch(main,args.num_gpus,num_machines=args.num_machines,machine_rank=args.machine_rank,dist_url=args.dist_url,args=(args,),)
