SSD#

With SSD, the following tasks are supported:

dataset_convert
train
evaluate
prune
inference
export

Data Input for Object Detection#

The object detection apps in TAO expect data in KITTI format for training and evaluation.

See the Data Annotation Format page for more information about the KITTI data format.

Pre-processing the Dataset#

The ssd dataloader supports the raw KITTI formatted data as well as TFrecords.

To use TFRecords for optimized iteration across the data batches, the the raw input data need to be converted to TFRecords format first. This can be done using the dataset_convert subtask. Currently, the KITTI and COCO formats are supported.

The dataset_convert tool requires a configuration file as input. Details of the configuration file and examples are included in the following sections.

Configuration File for Dataset Converter#

The dataset-converter specification describes how the corresponding data skill turns the original annotation format into the TFRecords format the trainer ingests. KITTI and COCO formats are configured via kitti_config or coco_config respectively. Use only one per specification file. The specification is a prototxt file with the following global parameters:

kitti_config: A nested prototxt configuration with multiple input parameters
coco_config: A nested prototxt configuration with multiple input parameters
image_directory_path: The path to the dataset root. The image_dir_name is appended to this path to get the input images and must be the same path specified in the experiment spec file.
target_class_mapping: The prototxt dictionary that maps the class names in the tfrecords to the target class to be trained in the network.

kitti_config#

Here are descriptions of the configurable parameters for the kitti_config field:

Parameter	Datatype	Default	Description	Supported Values
root_directory_path	string	–	The path to the dataset root directory	–
image_dir_name	string	–	The relative path to the directory containing images from the path in root_directory_path.	–
label_dir_name	string	–	The relative path to the directory containing labels from the path in root_directory_path.	–
partition_mode	string	–	The method employed when partitioning the data to multiple folds. Two methods are supported: Random partitioning: The data is divided in to 2 folds, train and val. This mode requires that the val_split parameter be set. Sequence-wise partitioning: The data is divided into n partitions (defined by the num_partitions parameter) based on the number of sequences available.	random sequence
num_partitions	int	2 (if partition_mode is random)	The number of partitions to use to split the data (N folds). This field is ignored when the partition model is set to random, as by default only two partitions are generated: val and train. In sequence mode, the data is split into n-folds. The number of partitions is ideally fewer than the total number of sequences in the kitti_sequence_to_frames file.	n=2 for random partition n< number of sequences in the kitti_sequence_to_frames_file
image_extension	str	.png	The extension of the images in the image_dir_name parameter.	.png .jpg .jpeg
val_split	float	20	The percentage of data to be separated for validation. This only works under “random” partition mode. This partition is available in fold 0 of the TFrecords generated. Set the validation fold to 0 in the dataset_config.	0-100
kitti_sequence_to_frames_file	str		The name of the KITTI sequence to frame mapping file. This file must be present within the dataset root as mentioned in the root_directory_path.
num_shards	int	10	The number of shards per fold.	1-20

The sample configuration file shown below converts the 100% KITTI dataset to the training set.

kitti_config {
  root_directory_path: "/workspace/tao-experiments/data/"
  image_dir_name: "training/image_2"
  label_dir_name: "training/label_2"
  image_extension: ".png"
  partition_mode: "random"
  num_partitions: 2
  val_split: 0
  num_shards: 10
}
image_directory_path: "/workspace/tao-experiments/data/"
  target_class_mapping {
      key: "car"
      value: "car"
  }
  target_class_mapping {
      key: "pedestrian"
      value: "pedestrian"
  }
  target_class_mapping {
      key: "cyclist"
      value: "cyclist"
  }
  target_class_mapping {
      key: "van"
      value: "car"
  }
  target_class_mapping {
      key: "person_sitting"
      value: "pedestrian"
  }
  target_class_mapping {
      key: "truck"
      value: "car"
  }

coco_config#

Here are descriptions of the configurable parameters for the coco_config field:

Parameter	Datatype	Default	Description	Supported Values
root_directory_path	string	–	The path to the dataset root directory	–
img_dir_names	string (repated)	–	The relative path to the directory containing images from the path in root_directory_path for each partition.	–
annotation_files	string (repated)	–	The relative path to the directory containing JSON file from the path in root_directory_path for each partition.	–
num_partitions	int	2	The number of partitions in the data. The number of partition must match the length of the list for img_dir_names and annotation_files. By default, two partitions are generated: val and train.
num_shards	int (repeated)	[10]	The number of shards per partitions. If only one value is provided, same number of shards is applied in all partitions

The sample configuration file shown below converts the COCO dataset with training and validation data where number of shard is 32 for validation and 256 for training.

coco_config {
  root_directory_path: "/workspace/tao-experiments/data/coco"
  img_dir_names: ["val2017", "train2017"]
  annotation_files: ["annotations/instances_val2017.json", "annotations/instances_train2017.json"]
  num_partitions: 2
  num_shards: [32, 256]
}
image_directory_path: "/workspace/tao-experiments/data/coco"

Sample Usage of the Dataset Converter Tool#

The dataset_convert tool is configured with a dataset spec describing the conversion from KITTI/COCO to TFRecord files.

Creating a Configuration File#

Below is a sample of the SSD specification file. It has six major components: ssd_config, training_config, eval_config, nms_config, augmentation_config, and dataset_config. The format of the specification file is a protobuf text(prototxt) message and each of its fields can be either a basic data type or a nested message.

random_seed: 42
ssd_config {
  aspect_ratios: "[[1.0, 2.0, 0.5],  [1.0, 2.0, 0.5, 3.0, 1.0/3.0],  [1.0, 2.0, 0.5, 3.0, 1.0/3.0],  [1.0, 2.0, 0.5, 3.0, 1.0/3.0],  [1.0, 2.0, 0.5],  [1.0, 2.0, 0.5]]"
  scales: "[0.07, 0.15, 0.33, 0.51, 0.69, 0.87, 1.05]"
  two_boxes_for_ar1: true
  clip_boxes: false
  variances: "[0.1, 0.1, 0.2, 0.2]"
  arch: "resnet"
  nlayers: 18
  freeze_bn: false
  freeze_blocks: 0
}
training_config {
  batch_size_per_gpu: 16
  num_epochs: 80
  enable_qat: false
  learning_rate {
  soft_start_annealing_schedule {
    min_learning_rate: 5e-5
    max_learning_rate: 2e-2
    soft_start: 0.15
    annealing: 0.8
    }
  }
  regularizer {
    type: L1
    weight: 3e-5
  }
}
eval_config {
  validation_period_during_training: 10
  average_precision_mode: SAMPLE
  batch_size: 16
  matching_iou_threshold: 0.5
}
nms_config {
  confidence_threshold: 0.01
  clustering_iou_threshold: 0.6
  top_k: 200
}
augmentation_config {
  output_width: 300
  output_height: 300
  output_channel: 3
  image_mean {
    key: 'b'
    value: 103.9
  }
  image_mean {
    key: 'g'
    value: 116.8
  }
  image_mean {
    key: 'r'
    value: 123.7
  }
}
dataset_config {
  data_sources: {
      # option 1
      tfrecords_path: "/path/to/train/tfrecord"

      # option 2
      # label_directory_path: "/path/to/train/labels"
      # image_directory_path: "/path/to/train/images"
  }
  include_difficult_in_training: true
  target_class_mapping {
      key: "car"
      value: "car"
  }
  target_class_mapping {
      key: "pedestrian"
      value: "pedestrian"
  }
  target_class_mapping {
      key: "cyclist"
      value: "cyclist"
  }
  target_class_mapping {
      key: "van"
      value: "car"
  }
  target_class_mapping {
      key: "person_sitting"
      value: "pedestrian"
  }
  validation_data_sources: {
    label_directory_path: "/path/to/val/labels"
    image_directory_path: "/path/to/val/images"
  }
}

The top level structure of the specification file is summarized in the sections below.

Training Config#

The training configuration (training_config) defines the parameters needed for the training, evaluation, and inference. Details are summarized in the table below.

Field	Description	Data Type and Constraints	Recommended/Typical Value
batch_size_per_gpu	The batch size for each GPU, so the effective batch size is batch_size_per_gpu * num_gpus	Unsigned int, positive	–
num_epochs	The number of epochs to train the network	Unsigned int, positive	–
enable_qat	Whether to use quantization-aware training	Boolean	Note: SSD does not support loading a pruned non-QAT model and retraining it with QAT enabled, or vice versa. For example, to get a pruned QAT model, perform the initial training with QAT enabled or `enable_qat=True`.
learning_rate	Only soft_start_annealing_schedule with these nested parameters is supported. 1. min_learning_rate: The minimum learning during the entire experiment 2. max_learning_rate: The maximum learning during the entire experiment 3. soft_start: Time to lapse before warm up ( expressed in percentage of progress between 0 and 1) 4. annealing: Time to start annealing the learning rate	Message type	–
regularizer	This parameter configures the regularizer to be used while training and contains the following nested parameters. 1. type: The type or regularizer to use. NVIDIA supports NO_REG, L1, and L2 2. weight: The floating point value for the regularizer weight	Message type	L1 Note: NVIDIA suggests using the L1 regularizer when training a network before pruning as L1 regularization helps make the network weights more prunable.
max_queue_size	The number of prefetch batches in data loading	Unsigned int, positive	–
n_workers	The number of workers for data loading (set to less than 4 when using tfrecords)	Unsigned int, positive	–
use_multiprocessing	Whether to use multiprocessing mode of keras sequence data loader	Boolean
visualizer	Training visualization config	Message type
early_stopping	Early stopping config	Message type

Note

The learning rate is automatically scaled with the number of GPUs used during training, or the effective learning rate is learning_rate * n_gpu.

Training Visualization Config#

Visualization during training is configured by the visualizer parameter. The parameters of it are described in the table below.

Parameter	Description	Data Type and Constraints	Recommended/Typical Value
enabled	Boolean flag to enable or disable this feature	bool.	–
num_images	The maximum number of images to be visualized in TensorBoard.	int.	`3`

If the visualization is enabled, the tensorboard log will be produced during training including the graphs for learning rate, training loss, validation loss, validation mAP and validation AP of each class. And the augmented images with bboxes will also be produced in the tensorboard.

Early Stopping#

The parameters for early stopping are described in the table below.

Parameter	Description	Data Type and Constraints	Recommended/Typical Value
monitor	The metric to monitor in order to enable early stopping.	string	`loss`
patience	The number of checks of `monitor` value before stopping the training.	int
min_delta	The delta of the minimum value of `monitor` value below which we regard it as not decreasing.	float

Evaluation Config#

The evaluation configuration (eval_config) defines the parameters needed for the evaluation either during training or as a standalone procedure. Details are summarized in the table below.

Field	Description	Data Type and Constraints	Recommended/Typical Value
validation_period_during_training	The number of training epochs per validation	Unsigned int, positive	10
average_precision_mode	The Average Precision (AP) calculation mode can be either SAMPLE or INTEGRATE. SAMPLE is used as VOC metrics for VOC 2009 or before. INTEGRATE is used for VOC 2010 or after.	ENUM type ( SAMPLE or INTEGRATE)	SAMPLE
matching_iou_threshold	The lowest IoU of the predicted box and ground truth box that can be considered a match.	Boolean	0.5
visualize_pr_curve	Boolean flag to enable or disable visualization of Precision-Recall curve.	Boolean

NMS Config#

The NMS configuration (nms_config) defines the parameters needed for NMS postprocessing. The NMS configuration applies to the NMS layer of the model in training, validation, evaluation, inference, and export. Details are summarized in the table below.

Field	Description	Data Type and Constraints	Recommended/Typical Value
confidence_threshold	Boxes with a confidence score less than confidence_threshold are discarded before applying NMS.	float	0.01
cluster_iou_threshold	The IoU threshold below which boxes will go through the NMS process.	float	0.6
top_k	top_k boxes will be output after the NMS keras layer. If the number of valid boxes is less than k, the returned array will be padded with boxes whose confidence score is 0.	Unsigned int	200
infer_nms_score_bits	The number of bits to represent the score values in NMS plugin in TensorRT OSS. The valid range is integers in [1, 10]. Setting it to any other values will make it fall back to ordinary NMS. Currently this optimized NMS plugin is only available in FP16 but it should also be selected by INT8 data type as there is no INT8 NMS in TensorRT OSS and hence this fastest implementation in FP16 will be selected. If falling back to ordinary NMS, the actual data type when building the engine will decide the exact precision(FP16 or FP32) to run at.	int. In the interval [1, 10].	0

Augmentation Config#

The augmentation_config parameter defines the image size after preprocessing. The augmentation methods in the SSD paper will be performed during training, including random flip, zoom-in, zoom-out and color jittering. And the augmented images will be resized to the output shape defined in augmentation_config. In evaluation process, only the resize will be performed.

Note

The details of augmentation methods can be found in setcion 2.2 and 3.6 of the paper.

Field	Description	Data Type and Constraints	Recommended/Typical Value
output_channel	Output image channel of augmentation pipeline.	integer	–
output_width	The width of preprocessed images and the network input.	integer, multiple of 32	–
output_height	The height of preprocessed images and the network input.	integer, multiple of 32	–
random_crop_min_scale	Minimum patch scale of RandomCrop augmentation. Default:0.3	float >= 1.0	–
random_crop_max_scale	Maximum patch scale of RandomCrop augmentation. Default:1.0	float >= 1.0	–
random_crop_min_ar	Minimum aspect ratio of RandomCrop augmentation. Default:0.5	float > 0	–
random_crop_max_ar	Maximum aspect ratio of RandomCrop augmentation. Default:2.0	float > 0	–
zoom_out_min_scale	Minimum scale of ZoomOut augmentation. Default:1.0	float >= 1.0	–
zoom_out_max_scale	Maximum scale of ZoomOut augmentation. Default:4.0	float >= 1.0	–
brightness	Brightness delta in color jittering augmentation. Default:32	integer >= 0	–
contrast	Contrast delta factor in color jitter augmentation. Default:0.5	float of [0, 1)	–
saturation	Saturation delta factor in color jitter augmentation. Default:0.5	float of [0, 1)	–
hue	Hue delta in color jittering augmentation. Default:18	integer >= 0	–
random_flip	Probablity of performing random horizontal flip. Default:0.5	float of [0, 1)	–
image_mean	A key/value pair to specify image mean values. If omitted, ImageNet mean will be used for image preprocessing. If set, depending on output_channel, either ‘r/g/b’ or ‘l’ key/value pair must be configured.	dict	–

Note

If set random_crop_min_scale = random_crop_max_scale = 1.0, RandomCrop augmentation will be disabled. Similarly, set zoom_out_min_scale = zoom_out_max_scale = 1, ZoomOut augmentation will be disabled. And all color jitter delta values are set to 0, color jittering augmentation will be disabled.

Dataset Config#

The dataset_config parameter defines the path to the training dataset, validation dataset, and target_class_mapping.

Field	Description	Data Type and Constraints	Recommended/Typical Value
data_sources	The path to the training dataset. When using tfrecord as dataset ingestion, set: tfrecords_path: The path to tfrecords When using raw KITTI labels and images, set: label_directory_path: The path to the label directory image_directory_path: The path to the image directory	Message type
include_difficult_in_training	Specifies whether to include difficult objects in the label (the Pascal VOC difficult label or KITTI occluded objects)	bool	true
validation_data_sources	The path to the training dataset images and labels	Message type
target_class_mapping	A mapping of classes in labels to the target classes	Message type

Note

data_sources and validation_data_sources are both repeated fields. Multiple datasets can be added to sources.

SSD config#

The SSD configuration (ssd_config) defines the parameters needed for building the SSD model. Details are summarized in the table below.

Field	Description	Data Type and Constraints	Recommended/Typical Value
aspect_ratios_global	The anchor boxes of aspect ratios defined in aspect_ratios_global will be generated for each feature layer used for prediction. Note that either the aspect_ratios_global or aspect_ratios parameter is required; you don’t need to specify both.	string	“[1.0, 2.0, 0.5, 3.0, 0.33]”
aspect_ratios	The aspect ratio of anchor boxes for different SSD feature layers Note: Either the aspect_ratios_global or aspect_ratios parameter is required; you don’t need to specify both.	string	“[[1.0,2.0,0.5], [1.0,2.0,0.5], [1.0,2.0,0.5], [1.0,2.0,0.5], [1.0,2.0,0.5], [1.0, 2.0, 0.5, 3.0, 0.33]]”
two_boxes_for_ar1	If this parameter is True, two boxes will be generated with an aspect ratio of 1. One with a scale for this layer and the other with a scale that is the geometric mean of the scale for this layer and the scale for the next layer.	Boolean	True
clip_boxes	If true, all corner anchor boxes will be truncated so they are fully inside the feature images.	Boolean	False
scales	A list of positive floats containing scaling factors per convolutional predictor layer. This list must be one element longer than the number of predictor layers so that, if two_boxes_for_ar1 is true, the second aspect-ratio 1.0 box for the last layer can have a proper scale. Except for the last element in this list, each positive float is the scaling factor for boxes in that layer. For example, if for one layer the scale is 0.1, then the generated anchor box with aspect ratio 1 for that layer (the first aspect-ratio 1 box if two_boxes_for_ar1 is set to True) will have its height and width as 0.1*min(img_h, img_w). min_scale and max_scale are two positive floats. If both of them appear in the config, the program can automatically generate the scales by evenly splitting the space between min_scale and max_scale.	string	“[0.05, 0.1, 0.25, 0.4, 0.55, 0.7, 0.85]”
min_scale/max_scale variances	If both appear in the config, scales will be generated evenly by splitting the space between min_scale and max_scale. A list of 4 positive floats. The four floats, in order, represent variances for box center x, box center y, log box height, and log box width. The box offset for box center (cx, cy) and log box size (height/width) w.r.t. anchor will be divided by their respective variance value. Therefore, larger variances result in less significant differences between two different boxes on encoded offsets.	float
steps	An optional list inside quotation marks with a length that is the number of feature layers for prediction. The elements should be floats or tuples/lists of two floats. The steps define how many pixels apart the anchor-box center points should be. If the element is a float, both vertical and horizontal margin is the same. Otherwise, the first value is step_vertical and the second value is step_horizontal. If steps are not provided, anchor boxes will be distributed uniformly inside the image.	string
offsets	An optional list of floats inside quotation marks with length equal to the number of feature layers for prediction. The first anchor box will have a margin of offsets[i]*steps[i] pixels from the left and top borders. If offsets are not provided, 0.5 will be used as default value.	string
arch	The backbone for feature extraction. Currently, “resnet”, “vgg”, “darknet”, “googlenet”, “mobilenet_v1”, “mobilenet_v2” and “squeezenet” are supported.	string	resnet
nlayers	The number of conv layers in a specific arch. For “resnet”, 10, 18, 34, 50 and 101 are supported. For “vgg”, 16 and 19 are supported. For “darknet”, 19 and 53 are supported. All other networks don’t have this configuration, and users should delete this parameter from the config file.	Unsigned int
freeze_bn	Whether to freeze all batch normalization layers during training.	boolean	False
freeze_blocks	The list of block IDs to be frozen in the model during training. You can choose to freeze some of the CNN blocks in the model to make the training more stable and/or easier to converge. The definition of a block is heuristic for a specific architecture. For example, by stride or by logical blocks in the model, etc. However, the block ID numbers identify the blocks in the model in a sequential order so you don’t have to know the exact locations of the blocks when you do training. As a general principle, the smaller the block ID, the closer it is to the model input; the larger the block ID, the closer it is to the model output. You can divide the whole model into several blocks and optionally freeze a subset of it. Note that for FasterRCNN, you can only freeze the blocks that are before the ROI pooling layer. Any layer after the ROI pooling layer will not be frozen anyway. For different backbones, the number of blocks and the block ID for each block are different. It deserves some detailed explanations on how to specify the block IDs for each backbone.	list(repeated integers) ResNet series. For the ResNet series, the block IDs valid for freezing is any subset of [0, 1, 2, 3] (inclusive) VGG series. For the VGG series, the block IDs valid for freezing is any subset of[1, 2, 3, 4, 5] (inclusive) GoogLeNet. For the GoogLeNet, the block IDs valid for freezing is any subset of[0, 1, 2, 3, 4, 5, 6, 7] (inclusive) MobileNet V1. For the MobileNet V1, the block IDs valid for freezing is any subset of [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11] (inclusive) MobileNet V2. For the MobileNet V2, the block IDs valid for freezing is any subset of [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13] (inclusive) DarkNet. For the DarkNet 19 and DarkNet 53, the block IDs valid for freezing is any subset of [0, 1, 2, 3, 4, 5] (inclusive)

Training the Model#

Input Requirement#

Input size: C * W * H (where C = 1 or 3, W >= 128, H >= 128)
Image format: JPG, JPEG, PNG
Label format: KITTI detection

Evaluating the Model#

Evaluation runs against the validation set specified in the training specification file.

Running Inference on the Model#

The inference command for SSD networks can be used to visualize bboxes or generate frame-by-frame KITTI format labels on a directory of images.

Pruning the Model#

Pruning removes parameters from the model to reduce the model size without compromising the integrity of the model itself.

After pruning, the model needs to be retrained. See Re-training the Pruned Model for more details.

Re-training the Pruned Model#

Once the model has been pruned, there might be a slight decrease in accuracy. This happens because some previously useful weights may have been removed. To regain accuracy, NVIDIA recommends that you retrain this pruned model over the same dataset with an updated specification file that points to the newly pruned model as the pretrained model file.

Users are advised to turn off the regularizer in the training_config for SSD to recover the accuracy when retraining a pruned model. You may do this by setting the regularizer type to NO_REG, as mentioned here. All the other parameters may be retained in the specification file from the previous training.

Note

SSD does not support loading a pruned non-QAT model and retraining it with QAT enabled, or vice versa. For example, to get a pruned QAT model, perform the initial training with QAT enabled or enable_qat=True.

Exporting the Model#

TAO includes the export command to export and prepare TAO models for Deploying to DeepStream. The export command optionally generates the calibration cache for TensorRT INT8 engine calibration.

Exporting the model decouples the training process from inference and allows conversion to TensorRT engines outside the TAO environment. TensorRT engines are specific to each hardware configuration and should be generated for each unique inference environment. This may be interchangeably referred to as the .trt or .engine file. The same exported TAO model may be used universally across training and deployment hardware. This is referred to as the .etlt file or encrypted TAO file. During model export, the TAO model is encrypted with a private key. This key is required when you deploy this model for inference.

INT8 Mode Overview#

TensorRT engines can be generated in INT8 mode to improve performance, but require a calibration cache at engine creation-time. The calibration cache is generated using a calibration tensor file, if export is run with the --data_type flag set to int8. Pre-generating the calibration information and caching it removes the need for calibrating the model on the inference machine. Moving the calibration cache is usually much more convenient than moving the calibration tensorfile, since it is a much smaller file and can be moved with the exported model. Using the calibration cache also speeds up engine creation as building the cache can take several minutes to generate depending on the size of the Tensorfile and the model itself.

The export tool can generate an INT8 calibration cache by ingesting training data using the following method:

Pointing the tool to a directory of images that you want to use to calibrate the model. For this option, make sure to create a sub-sampled directory of random images that best represent your training dataset.

FP16/FP32 Model#

The calibration.bin is only required if you need to run inference at INT8 precision. For FP16/FP32-based inference, the export step is much simpler: all you need to do is provide a model from the train step to export to convert it into an encrypted TAO model.

../../../../_images/fp16_fp32_export.png

QAT Export#

Note

When exporting a model that was trained with QAT enabled, the tensor scale factors to calibrate the activations are peeled out of the model and serialized to a TensorRT-readable cache file.

Deploying to DeepStream#

For deploying to deep stream, please refer to Deploying to DeepStream for SSD.