Hello iam using jetson inference to train object detection model. I did the training for 60 epochs and its complete.
Now i want to add more pictures to the dataset and retrain the model.
Is there a way to retrain the model rather than starting training from scratch?
Thank you
1 Like
Hi,
Yes. Please use the --pretrained-ssd
configuration to input a pre-trained model.
EX.
$ python3 train_ssd.py --pretrained-ssd=models/mobilenet-v1-ssd-mp-0_675.pth ...
For more configure support, you can check this file directly.
from vision.datasets.open_images import OpenImagesDataset
from vision.nn.multibox_loss import MultiboxLoss
from vision.ssd.config import vgg_ssd_config
from vision.ssd.config import mobilenetv1_ssd_config
from vision.ssd.config import squeezenet_ssd_config
from vision.ssd.data_preprocessing import TrainAugmentation, TestTransform
parser = argparse.ArgumentParser(
description='Single Shot MultiBox Detector Training With PyTorch')
# Params for datasets
parser.add_argument("--dataset-type", default="open_images", type=str,
help='Specify dataset type. Currently supports voc and open_images.')
parser.add_argument('--datasets', '--data', nargs='+', default=["data"], help='Dataset directory path')
parser.add_argument('--balance-data', action='store_true',
help="Balance training data by down-sampling more frequent labels.")
# Params for network
parser.add_argument('--net', default="mb1-ssd",
help="The network architecture, it can be mb1-ssd, mb1-lite-ssd, mb2-ssd-lite or vgg16-ssd.")
parser.add_argument('--freeze-base-net', action='store_true',
Thanks.
2 Likes