isaac_ros_tensor_proc

Source code available on GitHub.

API

Overview

The isaac_ros_tensor_proc package offers functionality for processing tensors, such as normalizing tensors, converting images to tensors, reshaping tensors and more. The main use case of this package is for preprocessing or postprocessing tensors following DNN model inference. One major use case of this package can be found in Isaac ROS DNN Image Encoder. The operations performed aim to be analogous to operations available in the PyTorch library.

Available Components

Component	Topics Subscribed	Topics Published	Parameters
ImageTensorNormalizeNode	tensor The input tensor. This should originate from an image	normalized_tensor The tensor that has been normalized	mean: The mean of the images per channel that will be used for normalization stddev: The standard deviation of the images per channel that will be used for normalization input_tensor_name: The name of the input tensor to extract output_tensor_name: The name of the output tensor to publish
ImageToTensorNode	image The input image.	tensor The resultant tensor, created using the input image	scale: Whether to scale the image by 255 or not tensor_name: The name of the output tensor to publish
InterleavedToPlanarNode	interleaved_tensor A tensor that’s in interleaved (e.g. HWC) format	planar_tensor A tensor that’s in planar (e.g. CHW) format	input_tensor_shape: The shape of the interleaved_tensor num_blocks: The number of blocks to preallocate output_tensor_name: The name of the output tensor to publish
NormalizeNode	image The input image.	normalized_tensor The tensor that has been created using the input image and then normalized.	image_mean: The mean of the images per channel that will be used for normalization image_stddev: The standard deviation of the images per channel that will be used for normalization input_image_width: The input image width input_image_height: The input image height num_blocks: The number of blocks to preallocate output_tensor_name: The name of the output tensor to publish
ReshapeNode	tensor The tensor whose dimensions will be reshaped	reshaped_tensor The reshaped tensor	output_tensor_name: The name of the output tensor to publish input_tensor_shape: The shape of the input tensor, tensor output_tensor_shape: The desired shape of the output tensor, reshaped_tensor num_blocks: The number of blocks to preallocate
TensorPairSyncNode	tensor1 The first input tensor tensor2 The second input tensor	tensor_pub The synchronized output containing both tensors	input_tensor1_name: The name of the first input tensor to extract input_tensor2_name: The name of the second input tensor to extract output_tensor1_name: The name for the first tensor in the output output_tensor2_name: The name for the second tensor in the output input_qos: QoS settings for input topics output_qos: QoS settings for output topics