print("Start create dataset!") # It will generate mindrecord file in args_opt.mindrecord_dir, # and the file name is yolo.mindrecord0, 1, ... file_num. if not os.path.isdir(args_opt.mindrecord_dir): os.makedirs(args_opt.mindrecord_dir) prefix = "yolo.mindrecord" mindrecord_file = os.path.join(args_opt.mindrecord_dir, prefix + "0") if not os.path.exists(mindrecord_file): if os.path.isdir(args_opt.image_dir) and os.path.exists(args_opt.anno_path): print("Create Mindrecord.") data_to_mindrecord_byte_image(args_opt.image_dir, args_opt.anno_path, args_opt.mindrecord_dir, prefix=prefix, file_num=8) print("Create Mindrecord Done, at {}".format(args_opt.mindrecord_dir)) else: print("image_dir or anno_path not exits.") if not args_opt.only_create_dataset: loss_scale = float(args_opt.loss_scale) # When create MindDataset, using the fitst mindrecord file, such as yolo.mindrecord0. dataset = create_yolo_dataset(mindrecord_file, repeat_num=args_opt.epoch_size, batch_size=args_opt.batch_size, device_num=device_num, rank=rank) dataset_size = dataset.get_dataset_size() print("Create dataset done!")
def main(): parser = argparse.ArgumentParser(description="YOLOv3 train") parser.add_argument("--only_create_dataset", type=bool, default=False, help="If set it true, only create " "Mindrecord, default is false.") parser.add_argument("--distribute", type=bool, default=False, help="Run distribute, default is false.") parser.add_argument("--device_id", type=int, default=0, help="Device id, default is 0.") parser.add_argument("--device_num", type=int, default=1, help="Use device nums, default is 1.") parser.add_argument("--lr", type=float, default=0.001, help="Learning rate, default is 0.001.") parser.add_argument("--mode", type=str, default="sink", help="Run sink mode or not, default is sink") parser.add_argument("--epoch_size", type=int, default=10, help="Epoch size, default is 10") parser.add_argument("--batch_size", type=int, default=32, help="Batch size, default is 32.") parser.add_argument("--checkpoint_path", type=str, default="", help="Checkpoint file path") parser.add_argument("--save_checkpoint_epochs", type=int, default=5, help="Save checkpoint epochs, default is 5.") parser.add_argument("--loss_scale", type=int, default=1024, help="Loss scale, default is 1024.") parser.add_argument( "--mindrecord_dir", type=str, default="./Mindrecord_train", help= "Mindrecord directory. If the mindrecord_dir is empty, it wil generate mindrecord file by" "image_dir and anno_path. Note if mindrecord_dir isn't empty, it will use mindrecord_dir " "rather than image_dir and anno_path. Default is ./Mindrecord_train") parser.add_argument("--image_dir", type=str, default="", help="Dataset directory, " "the absolute image path is joined by the image_dir " "and the relative path in anno_path") parser.add_argument("--anno_path", type=str, default="", help="Annotation path.") args_opt = parser.parse_args() context.set_context(mode=context.GRAPH_MODE, device_target="Ascend", device_id=args_opt.device_id) context.set_context(enable_loop_sink=True, enable_mem_reuse=True) if args_opt.distribute: device_num = args_opt.device_num context.reset_auto_parallel_context() context.set_auto_parallel_context( parallel_mode=ParallelMode.DATA_PARALLEL, mirror_mean=True, device_num=device_num) init() rank = args_opt.device_id % device_num else: rank = 0 device_num = 1 print("Start create dataset!") # It will generate mindrecord file in args_opt.mindrecord_dir, # and the file name is yolo.mindrecord0, 1, ... file_num. if not os.path.isdir(args_opt.mindrecord_dir): os.makedirs(args_opt.mindrecord_dir) prefix = "yolo.mindrecord" mindrecord_file = os.path.join(args_opt.mindrecord_dir, prefix + "0") if not os.path.exists(mindrecord_file): if os.path.isdir(args_opt.image_dir) and os.path.exists( args_opt.anno_path): print("Create Mindrecord.") data_to_mindrecord_byte_image(args_opt.image_dir, args_opt.anno_path, args_opt.mindrecord_dir, prefix=prefix, file_num=8) print("Create Mindrecord Done, at {}".format( args_opt.mindrecord_dir)) else: print("image_dir or anno_path not exits.") if not args_opt.only_create_dataset: loss_scale = float(args_opt.loss_scale) # When create MindDataset, using the fitst mindrecord file, such as yolo.mindrecord0. dataset = create_yolo_dataset(mindrecord_file, repeat_num=args_opt.epoch_size, batch_size=args_opt.batch_size, device_num=device_num, rank=rank) dataset_size = dataset.get_dataset_size() print("Create dataset done!") net = yolov3_resnet18(ConfigYOLOV3ResNet18()) net = YoloWithLossCell(net, ConfigYOLOV3ResNet18()) init_net_param(net, "XavierUniform") # checkpoint ckpt_config = CheckpointConfig(save_checkpoint_steps=dataset_size * args_opt.save_checkpoint_epochs) ckpoint_cb = ModelCheckpoint(prefix="yolov3", directory=None, config=ckpt_config) lr = Tensor( get_lr(learning_rate=args_opt.lr, start_step=0, global_step=args_opt.epoch_size * dataset_size, decay_step=1000, decay_rate=0.95, steps=True)) opt = nn.Adam(filter(lambda x: x.requires_grad, net.get_parameters()), lr, loss_scale=loss_scale) net = TrainingWrapper(net, opt, loss_scale) if args_opt.checkpoint_path != "": param_dict = load_checkpoint(args_opt.checkpoint_path) load_param_into_net(net, param_dict) callback = [ TimeMonitor(data_size=dataset_size), LossMonitor(), ckpoint_cb ] model = Model(net) dataset_sink_mode = False if args_opt.mode == "sink": print("In sink mode, one epoch return a loss.") dataset_sink_mode = True print( "Start train YOLOv3, the first epoch will be slower because of the graph compilation." ) model.train(args_opt.epoch_size, dataset, callbacks=callback, dataset_sink_mode=dataset_sink_mode)
device_id=args_opt.device_id) context.set_context(enable_task_sink=True, enable_loop_sink=True, enable_mem_reuse=True) config = ConfigSSD() prefix = "ssd_eval.mindrecord" mindrecord_dir = config.MINDRECORD_DIR mindrecord_file = os.path.join(mindrecord_dir, prefix + "0") if not os.path.exists(mindrecord_file): if not os.path.isdir(mindrecord_dir): os.makedirs(mindrecord_dir) if args_opt.dataset == "coco": if os.path.isdir(config.COCO_ROOT): print("Create Mindrecord.") data_to_mindrecord_byte_image("coco", False, prefix) print("Create Mindrecord Done, at {}".format(mindrecord_dir)) else: print("COCO_ROOT not exits.") else: if os.path.isdir(config.IMAGE_DIR) and os.path.exists( config.ANNO_PATH): print("Create Mindrecord.") data_to_mindrecord_byte_image("other", False, prefix) print("Create Mindrecord Done, at {}".format(mindrecord_dir)) else: print("IMAGE_DIR or ANNO_PATH not exits.") print("Start Eval!") ssd_eval(mindrecord_file, args_opt.checkpoint_path)
def main(): parser = argparse.ArgumentParser(description="SSD training") parser.add_argument("--only_create_dataset", type=bool, default=False, help="If set it true, only create " "Mindrecord, default is false.") parser.add_argument("--distribute", type=bool, default=False, help="Run distribute, default is false.") parser.add_argument("--device_id", type=int, default=0, help="Device id, default is 0.") parser.add_argument("--device_num", type=int, default=1, help="Use device nums, default is 1.") parser.add_argument("--lr", type=float, default=0.25, help="Learning rate, default is 0.25.") parser.add_argument("--mode", type=str, default="sink", help="Run sink mode or not, default is sink.") parser.add_argument("--dataset", type=str, default="coco", help="Dataset, defalut is coco.") parser.add_argument("--epoch_size", type=int, default=70, help="Epoch size, default is 70.") parser.add_argument("--batch_size", type=int, default=32, help="Batch size, default is 32.") parser.add_argument("--pre_trained", type=str, default=None, help="Pretrained Checkpoint file path.") parser.add_argument("--pre_trained_epoch_size", type=int, default=0, help="Pretrained epoch size.") parser.add_argument("--save_checkpoint_epochs", type=int, default=5, help="Save checkpoint epochs, default is 5.") parser.add_argument("--loss_scale", type=int, default=1024, help="Loss scale, default is 1024.") args_opt = parser.parse_args() context.set_context(mode=context.GRAPH_MODE, device_target="Ascend", device_id=args_opt.device_id) if args_opt.distribute: device_num = args_opt.device_num context.reset_auto_parallel_context() context.set_auto_parallel_context( parallel_mode=ParallelMode.DATA_PARALLEL, mirror_mean=True, device_num=device_num) init() rank = args_opt.device_id % device_num else: rank = 0 device_num = 1 print("Start create dataset!") # It will generate mindrecord file in args_opt.mindrecord_dir, # and the file name is ssd.mindrecord0, 1, ... file_num. config = ConfigSSD() prefix = "ssd.mindrecord" mindrecord_dir = config.MINDRECORD_DIR mindrecord_file = os.path.join(mindrecord_dir, prefix + "0") if not os.path.exists(mindrecord_file): if not os.path.isdir(mindrecord_dir): os.makedirs(mindrecord_dir) if args_opt.dataset == "coco": if os.path.isdir(config.COCO_ROOT): print("Create Mindrecord.") data_to_mindrecord_byte_image("coco", True, prefix) print("Create Mindrecord Done, at {}".format(mindrecord_dir)) else: print("COCO_ROOT not exits.") else: if os.path.isdir(config.IMAGE_DIR) and os.path.exists( config.ANNO_PATH): print("Create Mindrecord.") data_to_mindrecord_byte_image("other", True, prefix) print("Create Mindrecord Done, at {}".format(mindrecord_dir)) else: print("IMAGE_DIR or ANNO_PATH not exits.") if not args_opt.only_create_dataset: loss_scale = float(args_opt.loss_scale) # When create MindDataset, using the fitst mindrecord file, such as ssd.mindrecord0. dataset = create_ssd_dataset(mindrecord_file, repeat_num=args_opt.epoch_size, batch_size=args_opt.batch_size, device_num=device_num, rank=rank) dataset_size = dataset.get_dataset_size() print("Create dataset done!") ssd = SSD300(backbone=ssd_mobilenet_v2(), config=config) net = SSDWithLossCell(ssd, config) init_net_param(net) # checkpoint ckpt_config = CheckpointConfig(save_checkpoint_steps=dataset_size * args_opt.save_checkpoint_epochs) ckpoint_cb = ModelCheckpoint(prefix="ssd", directory=None, config=ckpt_config) if args_opt.pre_trained: if args_opt.pre_trained_epoch_size <= 0: raise KeyError( "pre_trained_epoch_size must be greater than 0.") param_dict = load_checkpoint(args_opt.pre_trained) load_param_into_net(net, param_dict) lr = Tensor( get_lr(global_step=args_opt.pre_trained_epoch_size * dataset_size, lr_init=0, lr_end=0, lr_max=args_opt.lr, warmup_epochs=max(350 // 20, 1), total_epochs=350, steps_per_epoch=dataset_size)) opt = nn.Momentum( filter(lambda x: x.requires_grad, net.get_parameters()), lr, 0.9, 0.0001, loss_scale) net = TrainingWrapper(net, opt, loss_scale) callback = [ TimeMonitor(data_size=dataset_size), LossMonitor(), ckpoint_cb ] model = Model(net) dataset_sink_mode = False if args_opt.mode == "sink": print("In sink mode, one epoch return a loss.") dataset_sink_mode = True print( "Start train SSD, the first epoch will be slower because of the graph compilation." ) model.train(args_opt.epoch_size, dataset, callbacks=callback, dataset_sink_mode=dataset_sink_mode)