sid1057

Hello again!

2022-06-28T00:00:00+00:00

Hello again I am returning to the blogging life of a programmer. Last year I was busy with my day job, doing cool things with trains, you know. But now I am creating my own startup in the field of dashcams, ADAS and self-driving. And I want to share with you some videos and notes about the adas system under development. To make it even cooler, I decided to test my ADAS system in one of the most fun racing games ever: Flatout 2. The main idea of the game is that accidents are encouraged, you get more nitro from them, and therefore it will be doubly interesting, because the algorithm will try to avoid accidents, unlike other participants in the race. The algorithm is still under development and I will try to create a video for each important step. I will post them twice a week: Monday and Friday. Some information will be private, because you know, I try to do it for commercial purposes. But I will try to make as many open source projects as possible. And the next two repositories that I'll publish are a client for getting controls and visuals data from classic racing games; and a repository for building a self-driving racing agent from the ADAS unit and some navigational abstraction. Here you can find my videos from the game: Later I will add my speech for important videos and make the material look like a lecture. And wish me luck!

GitHub Digest 2021-08-17

2021-08-17T00:00:00+00:00

`cppbestpractices`

Review: a good point to start writing high-quality cpp-code.

`applied-ml`

Review: a list of various papers/videos/posts on the application of machine learning in real companies for real problems.

`awesome-semantic-segmentation-pytorch`

Review: it could be very useful repo for quickly training some basic semantic segmentation model for your task in PyTorch.

`TCMR_RELEASE`

Review: looks like a very good human pose and shape estimation. Also gifs look funny to me.

`OpenBot`

Review: good start for your first garage (apartment) robot.

`lanenet-lane-detection-pytorch`

Review: yet another real-time lane detection. But the source is unfortunately fresh.

`fastseg`

Review: proof of concept that mobileV3+LR-ASPP can be used for AD semantic segmentation. Use the official TorchVision (0.9.0+) implementation, quantize it, convert to onnx and win comma.ai.

`MotionNet`

Review: yet another BEV lidar object detection model.

`publish-gitbook`

Review: github actions to automate gitbook routine.

`ResDepth`

Review: high quality DNN to refine depth from your aerial photo set.

`YOLOX`

Review: a repository for using yolo3-5 in your embedded hardware with TRT, onnx, light versions and more.

`onnxjs`

Review: the official way to run your onnx dnn in web browsers.

`webdnn`

Review: good solution to run your network onnx in different web browsers.

`Awesome-Lane-Detection`

Review: awesome list for awesome lane detection.

`self-adapting-confidence`

Review: good idea and good implementation of the confidence estimation for depth estimation. I think it's a very important area in the field of depth estimation.

`mono-uncertainty`

Review: another implementation of the confidence estimation for depth estimation. Looks pretty good.

`autonomous-car-chase`

Review: it's funny. Waiting for CARLA or BeamNG fork.

`awesome-vehicle-security`

Review: an awesome list for awesome but underrated self-driving cars topic.

`Real-ESRGAN`

Review: one of the best SOTA super resolution methods.

`DASR`

Review: another cool repo for blind super resolution.

`onnx-tensorrt`

Review: cool repo for onnx models into trt converting. Useful if you work with embedded and mobile devices.

`awesome-github-profile-readme`

Review: update your github profile now if you wanna be in trend.

`clearml`

Review: another ml workflow manager. Looks good and possibly looks better with AWS integration.

`drift_drl`

Review: finally.. tokyo drift in CARLA.

`LiDAR-Obstacle-Detection`

Review: a good start point for beginners to understand how you could detect obstacles using lidar point clouds.

`2D-Feature-Tracking`

Review: really useful comparison of differenet feature detection and matching methods in opencv. Even in 2k21 classical feature matching can still be useful in real tasks.

`jetson-packages-family`

Review: packages for jetsons. Useful.

`ml-roadmap`

Review: ml roadmap. Cool for juniors and enthusiasts.

`labelCloud`

Review: lightweight tool for 3D labelling.

`sofgan`

Review: some cool portrait generator.

`MotionPlanning`

Review: a good start point for path planning algorithms understanding.

`game-of-noise`

Review: interesting paper about advanced noise augmentation.

`DeepFaceLab`

Review: scary shit.

`yolo-online-learning`

Review: an approach of object detection online learning using lidar data, tracking and etc. I think it will be very popular soon in AD.

`dynamic-video-depth`

Review: looks like upgraded version of facebook consistent depth from google. Useful for moviemakers.

`RL-Gallery`

Review: an awesome list about RL.

GitHub Digest 2021-07-23

2021-07-23T00:00:00+00:00

`tkDNN`

Review: best solution to run yolov4 on your jetson device.

`Lidar_For_AD_references`

Review: yet another list, but already full of interesting paper and approaches to the lidar perception, special for AD, including SOTA.

`autonomous-car-2020`

Review: interesting and simple self-driving car pipeline for competition.

`LiDAR-MOS`

Review: looks like a very useful perception approach for self-driving car planning unit.

`osm2xodr`

Review: very useful repository if you want to develop your indie self-driving car in your city.

`MIT-Driverless-CV-TrainingInfra`

Review: I really enjoy exploring source codes for formula students driverless. A good example of a general model customization for a realworld challenge in a self-governing competition.

`fsoco-dataset`

Review: boost dataset for your formula student driverless team.

`Pose-Augmentation`

Review: waiting for the code for very exciting approach of data pose augmentation using DNN.

`pyodi`

Review: simple visualization for your object detection dataset.

`ovc`

Review: too old, but too interesting attempt to create fully completed hardware schema for various tasks of computer vision.

`Deep-Tutorials-for-PyTorch`

Review: looks like a very cool deep learning tutorial in pytorch, better than the old official tutorial.

`MotionBlur-detection-by-CNN`

Review: it could be useful for many real-life computer vision applications.

`CNN-DSO`

Review: too old approach of combining DSO with monocular depth estimation, but some companies are really still try to sell you this shit.

`awesome-ctf`

Review: it's still cool to be a hacker.

Github Digest 2021-07-15

2021-07-15T00:00:00+00:00

`Formula-Student-Driverless-Simulator`

Review: AirSim fork with formula student driverless track. Cool.

`lanenet-lane-detection`

Review: Real-time lane detection dnn.

`awesome-autonomous-vehicles`

Review: A paper list about AV.

`open-adas`

Review: Initiative project to create adas on jetson devices.

`hacker-roadmap`

Review: It's still cool to be hacker.

`SynLiDAR`

Review: Yet another synthetic lidar dataset.

`awesome-mobile-robotics`

Review: Yet another awesome list.

`pystemd`

Review: Cool python wrapper for systemd stuff. Could be useful.

`rtkbase`

Review: I think it's always cool to have frontend visualization for any unit in your system.

`tensorrtx`

Review: Awesome work for awesome things!!! Deserves a star!

`papers`

Review: Yet another paper list.

`awesome-hand-pose-estimation`

Review: Yet another awesome list.

`ros2_jetson`

Review: Official repository for using ROS2 on your jetson device if you think you need to use ROS on your jetson device...

`Anime-Girls-Holding-Programming-Books`

Review: The only thing you needed. (my favorite repo)

Video review: How TRI Trains Better Computer Vision Models with PD Synthetic Data

2021-07-15T00:00:00+00:00

Good interview on the use of synthetic data in computer vision problems for autonomous driving.

Key points:

Synthetics are extremely useful
Case on how the tracker learned to work in conditions of a large number of closed objects on the simulator
Simulators are photorealistic enough to give a boost, and in the case of heavy models without them it is almost impossible to collect the required amount of high-quality labeled data
Very cool case of how they used multi-tasking deep learning to improve semantics where they took most of the data from the simulator and estimated the depth in real data using monocular depth network
Idea: you can use real data only for validation, and depending on errors for some case / class, try to generate more synthetic data automatically to fix the error
Gaidon is cool

Midas comparison

2021-07-11T00:00:00+00:00

Midas comparison

Full text will be added later…

GitHub Digest 2021-07-08

2021-07-08T00:00:00+00:00

`KD_Lib`

Review: interesting. May be used for switching your VGG20000 network in production.

`DI-drive`

Review: waiting for next carla challenge.

`single-view-autocalib`

Review: useful for high distorsion lenses. In others keys just multiply vector by zero.

GitHub Digest 2021-06-29

2021-06-29T00:00:00+00:00

`awesome-AutoML`

Review: yet another list. About NAS etc. Interesting for research. Interesting for startups.

`Real-SR`

Review: very good super-resolution repository. Have to see deeper.

`lidar_course`

Review: good start point for learning how to use lidars.

GitHub Digest 2021-06-22

2021-06-22T00:00:00+00:00

`Workshop on Autonomous Driving at CVPR'21`

Review: must be seeing.

`awesome-engineering-management`

Review: managment stuff. List. Evening reading.

`xcit`

Review: looks beauty, waiting for +100500% accuracy.

`Drago Anguelov – Machine Learning for Autonomous Driving at Scale`

Review: from CVPR 2020 workshop. Interesting presentation about technology scaling.

GitHub Digest 2021-06-15

2021-06-15T00:00:00+00:00

`nvitop`

Review: ex of your girlfriend if you are nvidia-smi.

`awesome-self-supervised-learning`

Review: yet another paper list.

`lidar_cloud_to_image`

Review: great stuff for sensof fusion and monocular depth prediction dataset generation.

`wavelet-monodepth`

Review: seems interesting.

`Motion Prediction for Vulnerable Road Users`

Review: it's very good video for understanding the problem of human motion prediction for self-driving cars.

`modern-unix`

Review: GREAT. PERFECT. I USE IT AS START POINT FOR MY LINUX INSTALLATION. LAST NOT LEAST.