Hi, there!
We have this idea to perform character detection and recognition using DetectNet v2 18. The classes total will be around 35 classes (we excluded “o” from the classes).
Below are the specs of our training:
random_seed: 42
dataset_config {
data_sources {
tfrecords_path: "/workspace/tlt-experiments/tfrecords/OCR_trainval/*"
image_directory_path: "/workspace/tlt-experiments/data_kitti/training"
}
image_extension: "png"
target_class_mapping {
key: "a"
value: "a"
}
target_class_mapping {
key: "b"
value: "b"
}
target_class_mapping {
key: "c"
value: "c"
}
target_class_mapping {
key: "d"
value: "d"
}
target_class_mapping {
key: "e"
value: "e"
}
target_class_mapping {
key: "f"
value: "f"
}
target_class_mapping {
key: "g"
value: "g"
}
target_class_mapping {
key: "h"
value: "h"
}
target_class_mapping {
key: "i"
value: "i"
}
target_class_mapping {
key: "j"
value: "j"
}
target_class_mapping {
key: "k"
value: "k"
}
target_class_mapping {
key: "l"
value: "l"
}
target_class_mapping {
key: "m"
value: "m"
}
target_class_mapping {
key: "n"
value: "n"
}
target_class_mapping {
key: "p"
value: "p"
}
target_class_mapping {
key: "q"
value: "q"
}
target_class_mapping {
key: "r"
value: "r"
}
target_class_mapping {
key: "s"
value: "s"
}
target_class_mapping {
key: "t"
value: "t"
}
target_class_mapping {
key: "u"
value: "u"
}
target_class_mapping {
key: "v"
value: "v"
}
target_class_mapping {
key: "w"
value: "w"
}
target_class_mapping {
key: "x"
value: "x"
}
target_class_mapping {
key: "y"
value: "y"
}
target_class_mapping {
key: "z"
value: "z"
}
target_class_mapping {
key: "zero"
value: "zero"
}
target_class_mapping {
key: "one"
value: "one"
}
target_class_mapping {
key: "two"
value: "two"
}
target_class_mapping {
key: "three"
value: "three"
}
target_class_mapping {
key: "four"
value: "four"
}
target_class_mapping {
key: "five"
value: "five"
}
target_class_mapping {
key: "six"
value: "six"
}
target_class_mapping {
key: "seven"
value: "seven"
}
target_class_mapping {
key: "eight"
value: "eight"
}
target_class_mapping {
key: "nine"
value: "nine"
}
validation_fold: 0
}
augmentation_config {
preprocessing {
output_image_width: 832
output_image_height: 320
min_bbox_width: 1.0
min_bbox_height: 1.0
output_image_channel: 3
}
spatial_augmentation {
zoom_min: 1.0
zoom_max: 1.0
translate_max_x: 8.0
translate_max_y: 8.0
}
color_augmentation {
hue_rotation_max: 25.0
saturation_shift_max: 0.20000000298
contrast_scale_max: 0.10000000149
contrast_center: 0.5
}
}
postprocessing_config {
target_class_config {
key: "a"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "b"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "c"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "d"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "e"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "f"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "g"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "h"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "i"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "j"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "k"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "l"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "m"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "n"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "p"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "q"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "r"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "s"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "t"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "u"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "v"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "w"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "x"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "y"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "z"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "zero"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "one"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "two"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "three"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "four"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "five"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "six"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "seven"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "eight"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "nine"
value {
clustering_config {
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 0.0500000007451
minimum_bounding_box_height: 20
}
}
}
}
model_config {
pretrained_model_file: "/workspace/tlt-experiments/pretrained_resnet18/tlt_resnet18_detectnet_v2_v1/resnet18.hdf5"
num_layers: 18
use_batch_norm: true
activation {
activation_type: "relu"
}
objective_set {
bbox {
scale: 35.0
offset: 0.5
}
cov {
}
}
training_precision {
backend_floatx: FLOAT32
}
arch: "resnet"
}
evaluation_config {
validation_period_during_training: 10
first_validation_epoch: 1
minimum_detection_ground_truth_overlap {
key: "a"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "b"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "c"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "d"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "e"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "f"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "g"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "h"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "i"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "j"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "k"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "l"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "m"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "n"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "p"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "q"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "r"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "s"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "t"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "u"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "v"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "w"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "x"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "y"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "z"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "zero"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "one"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "two"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "three"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "four"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "five"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "six"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "seven"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "eight"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "nine"
value: 0.5
}
evaluation_box_config {
key: "a"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "b"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "c"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "d"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "e"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "f"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "g"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "h"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "i"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "j"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "k"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "l"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "m"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "n"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "p"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "q"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "r"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "s"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "t"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "u"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "v"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "w"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "x"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "y"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "z"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "zero"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "one"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "two"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "three"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "four"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "five"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "six"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "seven"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "eight"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "nine"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
average_precision_mode: INTEGRATE
}
cost_function_config {
target_classes {
name: "a"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "b"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "c"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "d"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "e"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "f"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "g"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "h"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "i"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "j"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "k"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "l"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "m"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "n"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "p"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "q"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "r"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "s"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "t"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "u"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "v"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "w"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "x"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "y"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "z"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "zero"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "one"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "two"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "three"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "four"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "five"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "six"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "seven"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "eight"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
target_classes {
name: "nine"
class_weight: 1.0
coverage_foreground_weight: 0.0500000007451
objectives {
name: "cov"
initial_weight: 1.0
weight_target: 1.0
}
objectives {
name: "bbox"
initial_weight: 10.0
weight_target: 10.0
}
}
enable_autoweighting: true
max_objective_weight: 0.999899983406
min_objective_weight: 9.99999974738e-05
}
training_config {
batch_size_per_gpu: 4
num_epochs: 150
learning_rate {
soft_start_annealing_schedule {
min_learning_rate: 5e-06
max_learning_rate: 5e-04
soft_start: 0.10000000149
annealing: 0.699999988079
}
}
regularizer {
type: L1
weight: 3.00000002618e-09
}
optimizer {
adam {
epsilon: 9.99999993923e-09
beta1: 0.899999976158
beta2: 0.999000012875
}
}
cost_scaling {
initial_exponent: 20.0
increment: 0.005
decrement: 1.0
}
checkpoint_interval: 10
}
bbox_rasterizer_config {
target_class_config {
key: "a"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "b"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "c"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "d"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "e"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "f"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "g"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "h"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "i"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "j"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "k"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "l"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "m"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "n"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "p"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "q"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "r"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "s"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "t"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "u"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "v"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "w"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "x"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "y"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "z"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "zero"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "one"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "two"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "three"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "four"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "five"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "six"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "seven"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "eight"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
target_class_config {
key: "nine"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 1.0
cov_radius_y: 1.0
bbox_min_radius: 1.0
}
}
deadzone_radius: 0.400000154972
}
This is my validation cost for each epoch:
Epoch 1/150
=========================
Validation cost: 0.000172
Mean average_precision (in %): 0.1380
class name average precision (in %)
------------ --------------------------
a 1.18977
b 0.000670547
c 0
d 0.0021093
e 0.146221
eight 1.59756
f 0
five 0.00110909
four 0.00154823
g 0.162082
h 0.000517966
i 0
j 0
k 0.0300903
l 0
m 0.0429045
n 0.00100402
nine 0.100871
one 0.16408
p 0
q 0
r 0.0691893
s 0.0273741
seven 0.6563
six 0.0282418
t 0
three 0.0752139
two 0.0660339
u 0
v 0
w 0.00405961
x 0
y 0
z 0
zero 0.462199
Epoch 11/150
=========================
Validation cost: 0.000042
Mean average_precision (in %): 11.8013
class name average precision (in %)
------------ --------------------------
a 61.8694
b 0
c 0
d 0
e 0
eight 33.213
f 0
five 22.6798
four 0
g 0
h 0
i 0
j 0
k 0
l 0
m 0
n 0
nine 26.7809
one 73.1699
p 0
q 0
r 0
s 0
seven 63.6912
six 21.5929
t 0
three 31.8103
two 38.5451
u 0
v 0
w 0
x 0
y 0
z 0
zero 39.692
Epoch 21/150
=========================
Validation cost: 0.000040
Mean average_precision (in %): 23.5290
class name average precision (in %)
------------ --------------------------
a 92.702
b 5.96042
c 15.1809
d 9.29706
e 0.93238
eight 65.9862
f 0.089329
five 53.0326
four 1.74415
g 5.21551
h 0.63429
i 0
j 6.71468
k 19.589
l 10.567
m 21.0408
n 19.3321
nine 28.913
one 84.6
p 5.905
q 3.15296
r 2.91622
s 4.20946
seven 88.9715
six 28.2543
t 10.5377
three 57.0328
two 80.3544
u 1.64927
v 0.463423
w 8.47342
x 12.7357
y 4.67456
z 0
zero 72.6517
Epoch 31/150
=========================
Validation cost: 0.000029
Mean average_precision (in %): 21.9510
class name average precision (in %)
------------ --------------------------
a 96.7257
b 8.406
c 9.75537
d 8.21308
e 0.285714
eight 54.6686
f 0
five 46.6173
four 0
g 1.28309
h 0.854701
i 0
j 0.404135
k 10.7206
l 0.269006
m 15.3079
n 8.26294
nine 28.2802
one 88.8418
p 1.63612
q 0
r 4.78647
s 4.20331
seven 96.1231
six 29.5714
t 13.8985
three 61.594
two 92.4074
u 0
v 0
w 1.83532
x 3.63711
y 3.25851
z 0.0757576
zero 76.3624
Epoch 41/150
=========================
Validation cost: 0.000029
Mean average_precision (in %): 21.5202
class name average precision (in %)
------------ --------------------------
a 96.9115
b 10.2614
c 9.32018
d 0.570437
e 0.757576
eight 53.9793
f 0.0744048
five 36.0021
four 0
g 0.0431406
h 0
i 0
j 0
k 10.53
l 0.0637959
m 6.9336
n 8.7828
nine 28.258
one 91.8441
p 5.51621
q 0
r 4.29824
s 8.92167
seven 93.7126
six 25.2564
t 11.4327
three 76.4001
two 95.3756
u 0
v 0
w 1.02339
x 0.351035
y 5.39916
z 0.506329
zero 70.6821
Epoch 51/150
=========================
Validation cost: 0.000029
Mean average_precision (in %): 20.7965
class name average precision (in %)
------------ --------------------------
a 98.2846
b 6.86915
c 3.72337
d 6.79137
e 0
eight 59.9766
f 0
five 38.9678
four 0
g 0
h 0
i 0
j 0
k 3.14123
l 0
m 5.30696
n 0
nine 30.0484
one 87.5423
p 2.58922
q 0
r 3.3287
s 0
seven 96.0294
six 30.3261
t 10.2118
three 78.2004
two 92.8943
u 0
v 0
w 0
x 0
y 1.26582
z 0
zero 72.3804
We found out that some of the classes don’t have good precision. And actually, based on the current epoch, the mAP is increasing. So my hypothesis is that DetectNet 18 is suitable for a limited number of classes and not suitable for character recognition. So, is this the correct assumption? Or do you have any suggestions of architecture that we may use for character recognition? I would appreciate any ideas.
Best regards,
Jeff