• Hardware nvidia A100
• Network Type TrafficCamNet Unprunned
We have been trying to re-engineer the trafficannet model with Tao tolking following Detectnet_2, our model has become much dumber instead of improving, could you give us a hand?
Regards
Can you share the training spec file? How many classes did you train?
And what does your training images and test images look like?
For trafficcamnet pretrained model mentioned in ngc model card, TrafficCamNet | NVIDIA NGC, this trafficCamNet v1.0 model was trained on a proprietary dataset with more than 3 million objects for car class and other classes. Most of the training dataset was collected and labeled in-house from several traffic cameras in a city in the US. The dataset also contains 40,000 images from a variety of dashcam to help with generalization and discrimination across classes. This content was chosen to improve accuracy of the object detection for images from a traffic cam at a traffic intersection.
This is the training spec file (TSF). We have nine classes that are showed in TFS mapping schema.
random_seed: 42
dataset_config {
data_sources {
tfrecords_path: "/workspace/training2/data/tfrecords/coco_train*"
image_directory_path: "/workspace/training2/data/images/train"
}
validation_data_source: {
tfrecords_path: "/workspace/training2/data/tfrecords/coco_val*"
image_directory_path: "/workspace/training2/data/images/val"
}
image_extension: "png"
target_class_mapping {
key: "car"
value: "car"
}
target_class_mapping {
key: "truck"
value: "truck"
}
target_class_mapping {
key: "person"
value: "person"
}
target_class_mapping {
key: "bigcaravan"
value: "bigcaravan"
}
target_class_mapping {
key: "motorcycle"
value: "motorcycle"
}
target_class_mapping {
key: "truckwobox"
value: "truckwobox"
}
target_class_mapping {
key: "carwtrailer"
value: "carwtrailer"
}
target_class_mapping {
key: "smalltruck"
value: "smalltruck"
}
target_class_mapping {
key: "caravan"
value: "caravan"
}
}
augmentation_config {
preprocessing {
output_image_width: 960
output_image_height: 544
min_bbox_width: 1.0
min_bbox_height: 1.0
output_image_channel: 3
}
spatial_augmentation {
hflip_probability: 0.5
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: "car"
value {
clustering_config {
clustering_algorithm: DBSCAN
dbscan_confidence_threshold: 0.9
coverage_threshold: 0.00499999988824
dbscan_eps: 0.20000000298
dbscan_min_samples: 1
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "truck"
value {
clustering_config {
clustering_algorithm: DBSCAN
dbscan_confidence_threshold: 0.9
coverage_threshold: 0.00499999988824
dbscan_eps: 0.15000000596
dbscan_min_samples: 1
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "person"
value {
clustering_config {
clustering_algorithm: DBSCAN
dbscan_confidence_threshold: 0.9
coverage_threshold: 0.00499999988824
dbscan_eps: 0.15000000596
dbscan_min_samples: 1
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "bigcaravan"
value {
clustering_config {
clustering_algorithm: DBSCAN
dbscan_confidence_threshold: 0.9
coverage_threshold: 0.00499999988824
dbscan_eps: 0.15000000596
dbscan_min_samples: 1
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "motorcycle"
value {
clustering_config {
clustering_algorithm: DBSCAN
dbscan_confidence_threshold: 0.9
coverage_threshold: 0.00499999988824
dbscan_eps: 0.15000000596
dbscan_min_samples: 1
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "truckwobox"
value {
clustering_config {
clustering_algorithm: DBSCAN
dbscan_confidence_threshold: 0.9
coverage_threshold: 0.00499999988824
dbscan_eps: 0.15000000596
dbscan_min_samples: 1
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "carwtrailer"
value {
clustering_config {
clustering_algorithm: DBSCAN
dbscan_confidence_threshold: 0.9
coverage_threshold: 0.00499999988824
dbscan_eps: 0.15000000596
dbscan_min_samples: 1
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "smalltruck"
value {
clustering_config {
clustering_algorithm: DBSCAN
dbscan_confidence_threshold: 0.9
coverage_threshold: 0.00499999988824
dbscan_eps: 0.15000000596
dbscan_min_samples: 1
minimum_bounding_box_height: 20
}
}
}
target_class_config {
key: "caravan"
value {
clustering_config {
clustering_algorithm: DBSCAN
dbscan_confidence_threshold: 0.9
coverage_threshold: 0.00499999988824
dbscan_eps: 0.15000000596
dbscan_min_samples: 1
minimum_bounding_box_height: 20
}
}
}
}
model_config {
pretrained_model_file: "/workspace/training2/experiments/pretrained_trafficcamnet/trafficcamnet_vunpruned_v1.0/resnet18_trafficcamnet.tlt"
num_layers: 18
use_batch_norm: true
objective_set {
bbox {
scale: 35.0
offset: 0.5
}
cov {
}
}
arch: "resnet"
}
evaluation_config {
validation_period_during_training: 1
first_validation_epoch: 50
minimum_detection_ground_truth_overlap {
key: "car"
value: 0.699999988079
}
minimum_detection_ground_truth_overlap {
key: "truck"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "person"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "bigcaravan"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "motorcycle"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "truckwobox"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "carwtrailer"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "smalltruck"
value: 0.5
}
minimum_detection_ground_truth_overlap {
key: "caravan"
value: 0.5
}
evaluation_box_config {
key: "car"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "truck"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "person"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "bigcaravan"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "motorcycle"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "truckwobox"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "carwtrailer"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "smalltruck"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
evaluation_box_config {
key: "caravan"
value {
minimum_height: 20
maximum_height: 9999
minimum_width: 10
maximum_width: 9999
}
}
average_precision_mode: INTEGRATE
}
cost_function_config {
target_classes {
name: "car"
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: "truck"
class_weight: 8.0
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: 1.0
}
}
target_classes {
name: "person"
class_weight: 8.0
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: 1.0
}
}
target_classes {
name: "bigcaravan"
class_weight: 8.0
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: 1.0
}
}
target_classes {
name: "motorcycle"
class_weight: 8.0
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: 1.0
}
}
target_classes {
name: "truckwobox"
class_weight: 8.0
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: 1.0
}
}
target_classes {
name: "carwtrailer"
class_weight: 8.0
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: 1.0
}
}
target_classes {
name: "smalltruck"
class_weight: 8.0
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: 1.0
}
}
target_classes {
name: "caravan"
class_weight: 8.0
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: 1.0
}
}
enable_autoweighting: false
max_objective_weight: 0.999899983406
min_objective_weight: 9.99999974738e-05
}
training_config {
batch_size_per_gpu: 32
num_epochs: 500
learning_rate {
soft_start_annealing_schedule {
min_learning_rate: 5e-07
max_learning_rate: 1e-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
}
visualizer{
enabled: true
num_images: 9
scalar_logging_frequency: 5
infrequent_logging_frequency: 5
target_class_config {
key: "car"
value: {
coverage_threshold: 0.005
}
}
target_class_config {
key: "truck"
value: {
coverage_threshold: 0.005
}
}
target_class_config {
key: "person"
value: {
coverage_threshold: 0.005
}
}
target_class_config {
key: "bigcaravan"
value: {
coverage_threshold: 0.005
}
}
target_class_config {
key: "motorcycle"
value: {
coverage_threshold: 0.005
}
}
target_class_config {
key: "truckwobox"
value: {
coverage_threshold: 0.005
}
}
target_class_config {
key: "carwtrailer"
value: {
coverage_threshold: 0.005
}
}
target_class_config {
key: "smalltruck"
value: {
coverage_threshold: 0.005
}
}
target_class_config {
key: "caravan"
value: {
coverage_threshold: 0.005
}
}
clearml_config{
project: "TAO Toolkit ClearML Demo"
task: "detectnet_v2_resnet18_clearml"
tags: "detectnet_v2"
tags: "training"
tags: "resnet18"
tags: "unpruned"
}
wandb_config{
project: "VISTA"
name: "trafficcamnet"
tags: "detectnet_v2"
tags: "training"
tags: "resnet18"
tags: "unpruned"
}
}
checkpoint_interval: 10
}
bbox_rasterizer_config {
target_class_config {
key: "car"
value {
cov_center_x: 0.5
cov_center_y: 0.5
cov_radius_x: 0.40000000596
cov_radius_y: 0.40000000596
bbox_min_radius: 1.0
}
}
target_class_config {
key: "truck"
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: "person"
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: "bigcaravan"
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: "motorcycle"
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: "truckwobox"
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: "carwtrailer"
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: "smalltruck"
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: "caravan"
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
}
Our dataset’s images have an high altitude point of view. Scene focus is on the vehicles that comes on each road line.
OK, this may be different from the training images of trafficcamnet since they are from several traffic cameras in a city.
From an high altitude point of view, some of objects may be small or even less than 16x16. As mentioned in ngc website, this is the limitation of trafficcamnet(i.e, detectnet_v2 nework).
The 9 classes are also more than the trafficcamnet. Suggest you to run YOLO_v4_tiny network in TAO. YOLOv4_tiny network can have a better detection against smaller objects and more classes. Please train against your own custom annotated dataset, with ngc object detection pretrained model(hdf5 format). More info can be found in tao_tutorials/notebooks/tao_launcher_starter_kit/yolo_v4_tiny at main · NVIDIA/tao_tutorials · GitHub.
Hello,
We believe that this is the problem we have, we have created another new thread, I’m telling you to close this one