Please provide the following information when requesting support.
• Hardware: Gefore 1050ti
• Network Type: LPRnet
• TLT Version: http://nvcr.io/nvidia/tao/tao-toolkit-tf:v3.21.08-py3
• Training spec file
random_seed: 42
lpr_config {
hidden_units: 512
max_label_length: 9
arch: "baseline"
nlayers: 18 #setting nlayers to be 10 to use baseline10 model
}
training_config {
batch_size_per_gpu: 32
num_epochs: 5
learning_rate {
soft_start_annealing_schedule {
min_learning_rate: 1e-6
max_learning_rate: 1e-4
soft_start: 0.001
annealing: 0.7
}
}
regularizer {
type: L2
weight: 5e-4
}
}
eval_config {
validation_period_during_training: 5
batch_size: 1
}
augmentation_config {
output_width: 96
output_height: 48
output_channel: 3
max_rotate_degree: 5
rotate_prob: 0.5
keep_original_prob: 0.3
}
dataset_config {
data_sources: {
label_directory_path: '/workspace/tlt-experiments/licenseplate_dataset_lprnet/train_Lao_VN_v4/train/labels'
image_directory_path: '/workspace/tlt-experiments/licenseplate_dataset_lprnet/train_Lao_VN_v4/train/images'
}
characters_list_file: '/workspace/tlt-experiments/lprnet/specs/lao_vn_lp_characters.txt'
validation_data_sources: {
label_directory_path: '/workspace/tlt-experiments/licenseplate_dataset_lprnet/Lao_LPR_v2_Split/valid/labels'
image_directory_path: '/workspace/tlt-experiments/licenseplate_dataset_lprnet/Lao_LPR_v2_Split/valid/images'
}
}
• How to reproduce the issue ? (This is for errors. Please share the command line and the detailed log here.)
Given a license plate image, LPRNet should recognize all the characters in the image. For example, “43A 123456”
However, after integrating the LPRNet to the Deepstream, the result was 43A 12345. The number 6 was disappeared.
This did not happen when using the LPRNet standalone.
This phenomenon only happened several times.
The application is carried out in two phases:
- License Plate Detection → License Plate Recognition.
Real Example
The image below is the a real image
- Model prediction: 37C1954
- Expected Result: 37C19545