2.0
Introduction
Overview
NVIDIA Transfer Learning Toolkit
Pre-trained Models
TLT Workflow Overview
Requirements and Installation
Hardware Requirements
Minimum
Recommended
Software Requirements
Installation Prerequisites
Installation
Running the Transfer Learning Toolkit
Run the toolkit
Access local directories
Use the examples
Downloading the Models
Configure the NGC API key
Get a list of models
Download a model
Supported Model Architectures
Model Requirements
Classification
Object Detection
DetectNet_v2
FasterRCNN
SSD
DSSD
YOLOv3
RetinaNet
Instance Segmentation
MaskRCNN
Training
Deployment
Purpose-Built Models
Training
Deployment
TrafficCamNet
PeopleNet
DashCamNet
FaceDetect-IR
VehicleMakeNet
VehicleTypeNet
QAT and AMP for Training
TLT Workflow
Augmenting a Dataset
Configuring the Augmentor
Spatial Augmentation Config
Rotation Config
Shear Config
Flip Config
Translation Config
Color Augmentation Config
Hue Saturation Config
Brightness Config
Contrast Config
Dataloader
Blur Config
Running the Augmentor Tool
Preparing the Input Data Structure
Data Input for Classification
Data Input for Object Detection
KITTI file format
Label Files
Sequence Mapping File
Conversion to TFRecords
Configuration File for Dataset Converter
Sample Usage of the Dataset Converter Tool
Data Input for Instance Segmentation
COCO format for Instance Segmentation
Creating an Experiment Spec File
Specification File for Classification
Model Config
Eval Config
Training Config
Specification File for DetectNet_v2
Model Config
BBox Ground Truth Generator
Post processor
Cost Function
Trainer
Augmentation Module
Configuring the Evaluator
Dataloader
Specification File for Inference
Inferencer
TLT_Config
Bbox Handler
Specification File for FasterRCNN
Network Config
Training Configuration
Specification File for SSD
Training Config
Evaluation Config
NMS Config
Augmentation Config
Dataset Config
SSD config
Specification File for DSSD
Training Config
Evaluation Config
NMS Config
Augmentation Config
Dataset Config
DSSD Config
Using aspect_ratios_global or aspect_ratios
two_boxes_for_ar1
scales or Combination of min_scale and max_scale
clip_boxes
loss_loc_weight
focal_loss_alpha and focal_loss_gamma
variances
steps
offsets
arch
nlayers
freeze_bn
freeze_blocks
Specification File for RetinaNet
Training Config
Evaluation Config
NMS Config
Augmentation Config
Dataset Config
RetinaNet Config
Specification File for YOLOv3
Training Config
Evaluation Config
NMS Config
Augmentation Config
Dataset Config
YOLOv3 Config
Specification File for MaskRCNN
MaskRCNN Config
Data Config
Training the Model
Quantization Aware Training
Automatic Mixed Precision
Training a classification model
Required Arguments
Optional Arguments
Training a DetectNet_v2 model
Required Arguments
Optional Arguments
Sample Usage
Training a FasterRCNN Model
Required Arguments
Optional Arguments
Sample Usage
Using a Pretrained Weights File
Training an SSD Model
Required Arguments
Optional Arguments
Sample Usage
Training a DSSD Model
Required arguments
Optional Arguments
Sample Usage
Training a YOLOv3 Model
Required Arguments
Optional Arguments
Sample Usage
Training a RetinaNet model
Required Arguments
Optional Arguments
Sample Usage
Training a MaskRCNN Model
Required Arguments
Optional Arguments
Sample Usage
Evaluating the Model
Evaluating a Classification Model
Required Arguments
Optional Arguments
Evaluating a DetectNet_v2 Model
Required Arguments
Optional Arguments
Evaluating a FasterRCNN Model
Required Arguments
Optional Arguments
Evaluation Metrics
Two Modes for tlt-evaluate
Evaluating an SSD Model
Required Arguments
Optional Arguments
Evaluating a DSSD model
Required Arguments
Optional Arguments
Evaluating a YOLOv3 Model
Required Arguments
Optional Arguments
Evaluating a RetinaNet Model
Required Arguments
Optional Arguments
Evaluating a MaskRCNN Model
Required Arguments
Optional Arguments
Using Inference on a Model
Running Inference on a Classification Model
Required arguments
Optional arguments
Running Inference on a DetectNet_v2 Model
Required Parameters
Running Inference on a FasterRCNN Model
Required Arguments
Optional Arguments
Two Modes for tlt-infer
Running Inference on an SSD Model
Required Arguments
Optional Arguments
Running Inference on a DSSD Model
Required Arguments
Optional Arguments
Running Inference on a YOLOv3 Model
Required Arguments
Optional Arguments
Running Inference on a RetinaNet Model
Required Arguments
Optional Arguments
Required Arguments
Optional Arguments
Required Arguments
Optional arguments
Running Inference on a MaskRCNN Model
Required Arguments
Optional Arguments
Required Arguments
Optional Arguments
Pruning the Model
Required Arguments
Optional Arguments
Using the Prune Command
Re-training the Pruned Model
Exporting the Model
INT8 Mode Overview
FP16/FP32 Model
Generating an INT8 tensorfile Using the tlt-int8-tensorfile Command
Positional Arguments
Required Arguments
Optional Argument
Exporting the Model Using tlt-export
Required Arguments
Optional Arguments
INT8 Export Mode Required Arguments
INT8 Export Optional Arguments
Exporting a Model
Deploying to Deepstream
TensorRT Open Source Software (OSS)
TensorRT OSS on x86
TensorRT OSS on Jetson (ARM64)
Generating an Engine Using tlt-converter
Instructions for x86
Instructions for Jetson
Using the tlt-converter
Required Arguments
Optional Arguments
INT8 Mode Arguments
Sample Output Log
Integrating the model to DeepStream
Integrating a Classification model
Label File
DeepStream Configuration File
Integrating a DetectNet_v2 model
Label File
DeepStream Configuration File
Integrating an SSD Model
Prerequisites for SSD Model
Label File
DeepStream Configuration File
Integrating a FasterRCNN Model
Prerequisite for FasterRCNN Model
Label File
DeepStream Configuration File
Integrating a YOLOv3 Model
Prerequisite for YOLOv3 model
Label File
DeepStream Configuration File
Integrating a DSSD Model
Prerequisite for DSSD model
Label File
DeepStream Configuration File
Integrating a RetinaNet Model
Prerequisite for RetinaNet Model
Label File
DeepStream Configuration File
Integrating Purpose-Built Models
Integrating a MaskRCNN Model
deepstream-app Config File
Label File
More Information
Release Notes
Transfer Learning Toolkit for Intelligent Video Analytics V2.0
Key Features
Contents
Software Requirements
Hardware Requirements
Minimum
Recommended
Known Issues
Resolved Issues
Frequently Asked Questions
Model support
Pruning
Model Export and Deployment
Training
Transfer Learning Toolkit
Docs
»
QAT and AMP for Training
View page source
QAT and AMP for Training
ΒΆ