Data

Tools

Indexes

Applications

Experiments

Open Source Science

depthstream-accelerator-ros2-integrated-monocular-depth-inference

DepthStream Accelerator: A TensorRT-optimized monocular depth estimation tool with ROS2 integration for C++. It offers high-speed, accurate depth perception, perfect for real-time applications in robotics, autonomous vehicles, and interactive 3D environments.

https://github.com/jagennath-hari/depthstream-accelerator-ros2-integrated-monocular-depth-inference

Science Score: 44.0%

This score indicates how likely this project is to be science-related based on various indicators:

  • ✓
    CITATION.cff file
    Found CITATION.cff file
  • ✓
    codemeta.json file
    Found codemeta.json file
  • ✓
    .zenodo.json file
    Found .zenodo.json file
  • ○
    DOI references
  • ○
    Academic publication links
  • ○
    Academic email domains
  • ○
    Institutional organization owner
  • ○
    JOSS paper metadata
  • ○
    Scientific vocabulary similarity
    Low similarity (11.7%) to scientific vocabulary

Keywords

computer-vision deep-learning depth-estimation inference-acceleration monocular-depth-estimation robotics ros2 tensorrt-inference vision-tranformer
Last synced: 6 months ago · JSON representation ·

Repository

DepthStream Accelerator: A TensorRT-optimized monocular depth estimation tool with ROS2 integration for C++. It offers high-speed, accurate depth perception, perfect for real-time applications in robotics, autonomous vehicles, and interactive 3D environments.

Basic Info
  • Host: GitHub
  • Owner: jagennath-hari
  • License: bsd-3-clause
  • Language: C++
  • Default Branch: main
  • Homepage:
  • Size: 10.9 MB
Statistics
  • Stars: 17
  • Watchers: 1
  • Forks: 0
  • Open Issues: 0
  • Releases: 1
Topics
computer-vision deep-learning depth-estimation inference-acceleration monocular-depth-estimation robotics ros2 tensorrt-inference vision-tranformer
Created over 2 years ago · Last pushed 11 months ago
Metadata Files
Readme License Citation

README.md

DepthStream-Accelerator-ROS2-Integrated-Monocular-Depth-Inference

DepthStream Accelerator: A TensorRT-optimized monocular depth estimation tool with ROS2 integration for C++. It offers high-speed, accurate depth perception, perfect for real-time applications in robotics, autonomous vehicles, and interactive 3D environments.

🏁 Dependencies

1) NVIDIA Driver (Official Link) 2) CUDA Toolkit (Official Link) 3) cuDNN (Official Link) 4) TensorRT (Official Link) 5) OpenCV CUDA (Github Guide) 6) Miniconda (Official Link) 7) ROS 2 Humble (Official Link) 8) ZoeDepth (Official Link)

⚙️ Creating the Engine File

ONNX File

Create the onnx file cd monocular_depth/scripts/ZoeDepth/ && python trt_convert.py. The ONNX file gets saved in the working directory as zoe_nk.onnx, configure the input dimensions as per your input image dimensions (h, w).

You can download a prebuilt .onnx file from LINK.

TensorRT engine creation

Once you have the .onnx file created go into the tensorRT trtexec directory. Mostly this is cd /usr/src/tensorrt/bin/. Now it is time to create the engine file, this could take a few minutes to create. Run the command below, ./trtexec --onnx=zoe_nk.onnx --builderOptimizationLevel=3 --useSpinWait --useRuntime=full --useCudaGraph --precisionConstraints=obey --allowGPUFallback --tacticSources=+CUBLAS,+CUDNN,+JIT_CONVOLUTIONS,+CUBLAS_LT --inputIOFormats=fp32:chw --outputIOFormats=fp32:chw --sparsity=enable --layerOutputTypes=fp32 --layerPrecisions=fp32 --saveEngine=zoe_nk.trt

🖼️ Running Depth Estimation

Build the ROS2 workspace

colcon build --symlink-install --cmake-args=-DCMAKE_BUILD_TYPE=Release --parallel-workers $(nproc)

Inference

ros2 launch monocular_depth mono_depth.launch.py trt_path:=zoe_nk.trt image_topic:=/rgb/image_rect_color gui:=true With GUI.

ros2 launch monocular_depth mono_depth.launch.py trt_path:=zoe_nk.trt image_topic:=/rgb/image_rect_color gui:=false Without GUI.

GUI

SLAM

DEPTH MAP

💬 ROS2 Message

The depth map gets published as sensor_msg/Image in the /mono_depth/depthMap topic

SLAM

⚠️ Note

1) There is a know problem with MiDAS which doesnt allow tracing the network. 2) If the input image dimensions are changed, change the source code to match your image dimension inputs.

Owner

  • Name: Jagennath Hari
  • Login: jagennath-hari
  • Kind: user

Citation (CITATION.cff) 

cff-version: 1.2.0
message: "If you use this software, please cite it as below."
authors:
- family-names: "Hari"
  given-names: "Jagennath"
title: "DepthStream-Accelerator-ROS2-Integrated-Monocular-Depth-Inference"
version: 1.0.0
date-released: 2023-11-10
url: "https://github.com/jagennath-hari/DepthStream-Accelerator-ROS2-Integrated-Monocular-Depth-Inference"

GitHub Events

Total
  • Watch event: 8
  • Push event: 5
Last Year
  • Watch event: 8
  • Push event: 5