Posts

  • GDB - To debug your C++ ROS code more efficient

    Tue, Jan 18, 2022 - 200 Words
    In general, gbd is a very powerful tool for debugging C++ at runtime. In order to use it as you are developing your C++ ROS code, you need to do some extra steps. I actually learned this trick from a colleague, and want to save it for future use by writing this post. Step 1: Compile your code in debug mode. catkin build your_ros_pkg -DCMAKE_BUILD_TYPE=Debug Step 2: Run your node with gdb. Read more →

  • Inference on edge: tensorflow lite C++ and GPU/CPU/Hexagon DSP

    Sun, Mar 14, 2021 - 1000 Words
    Introduction As one of 10 winning startups of the Qualcomm Vietnam Innovative Challenge 2020, I have the chance to use the Qualcomm RB5 board to deploy our autonomous driving system for Automated Guided Vehicles (AGV). If you don’t know about AGV and the challenge, you can quickly found a good explanation using Google search. The main reason which urges me to write this post is that there is a limited number of tutorials you can found online on how to use TensorFlow lite C++ even the homepage of the TensorFlow package. Read more →

  • Make a conversation between two ESP32 using MQTT protocol

    Mon, Jan 4, 2021 - 800 Words
    Make a conversation between two ESP32 using MQTT protocol Introduction I got two ESP32 from a China shop for just $10 including the shipping fee to the US. ESP32 is a very powerful computer with a dual-core Xtensa® 32-bit LX6 240MHz MCU, even more powerful than a supercomputer in the 90s. Also, 448 KB of ROM and 520 KB of on-chip SRAM is abundant for embedded programs (mine are ESP32-WROOM-32D). Read more →

  • A single framework for inference from different pre-trained models (Caffe, MXnet, Darknet, Tensorflow, ...)

    Wed, Dec 16, 2020 - 800 Words
    Object Detection and Localization from overhead cameras in fixed area This micropackage includes: object_detector.py: A wraper of opencv dnn module for different sources of pre-trained models: caffe, tensorflow or darknet … You can use the same framework, which contains only 2 methods: forward() and post_process(), for all models from different architectures. localizer.py: A simple localization algorithm for overhead camera in fixed area You can: load the interested model in and use it directly to infer from an input. Read more →

Social media