profile
viewpoint
If you are wondering where the data of this site comes from, please visit https://api.github.com/users/cmpute/events. GitMemory does not store any data, but only uses NGINX to cache data for a period of time. The idea behind GitMemory is simply to give users a better reading experience.
Jacob Zhong cmpute University of Michigan Ann Arbor zyxin.xyz (o゚v゚)ノ

cmpute/awesome-uwp-apps 64

A list about open-sourced repositories of Win 10 (UWP) Applications in Github | In store or off store

cmpute/d3d 19

Devkit for 3D -- Some utils for 3D object detection based on Numpy and Pytorch

cmpute/chainer-regnet 10

Chainer implementation of RegNet: Multimodal sensor registration using deep neural networks (with simplifications)

cmpute/dgal 3

Differentiable Geometry Algorithms Library

cmpute/audio-codec-benchmark 2

Detailed comparison of lossless and lossy audio codecs

cmpute/cgal.py 2

Pybind11 binding of cgal. Aimed at extending and easy use

cmpute/chainer-voxelnet 1

VoxelNet implemented in chainer

CaoZhong1992/carla-challenge-route 0

carla-challenge-route

cmpute/15-minute-apps 0

15 minute (small) desktop apps built with PyQt

cmpute/AB3DMOT 0

Official Python Implementation for "3D Multi-Object Tracking: A Baseline and New Evaluation Metrics", IROS 2020, ECCVW 2020

startedpierrechevalier83/ferris

started time in 7 days

push eventcmpute/d3d

Jacob Zhong

commit sha b592e71a5d63939a2c5ea236d7aaf54b37b415f4

Add label mapping for SemanticKITTI

view details

push time in 10 days

startedeschnett/zotero-citationcounts

started time in 10 days

startedtuplex/tuplex

started time in 11 days

push eventcmpute/d3d

Jacob Zhong

commit sha fac03d6a1e4c1042dfcbdf4572867fc07228faf4

Update README.md

view details

push time in 13 days

startedholynova/gushi_namer

started time in 14 days

startedpatriksvensson/cupboard

started time in 14 days

startedventusff/neurecon

started time in 14 days

startedgoogle-research/ott

started time in 14 days

startedhyliang96/books

started time in 15 days

push eventcmpute/d3d

Jacob Zhong

commit sha 6610257b99c40c39ea89e52f0b36709191a218d3

Add detection and segmentation dumping in ros

view details

Jacob Zhong

commit sha 9a3fb8f095bd51957aed4fd58640842962828969

Minor fix

view details

push time in 16 days

push eventcmpute/pcl.py

Jacob Zhong

commit sha 1ee5f86aab4defa82e8bf6756fe43bad743aff8e

Improve append fields

view details

push time in 17 days

push eventcmpute/d3d

Jacob Zhong

commit sha 39dc916450d645da5551ba7762b3e2192f4e5ba0

Update dataset readme

view details

push time in 17 days

issue closedmpitropov/cadc_devkit

Unable to align the point clouds between two frames

Hi! Thanks for your amazing work on this dataset. I wanted to align two point clouds together with the given sensor intrinsics and novatel data. However the best I can do right now is illustrated below, in which there are still noticeable error between two frames (these two are 25 frames apart).

2021-08-31_15-07

Currently how I parse the pose data is by the following code

import utm
from scipy.spatial.transform import Rotation

def parse_pose_from_inspvax(data: INSPVAX):
    x, y, *_ = utm.from_latlon(data.latitude, data.longitude)
    t = [-x, -y, data.altitude + data.undulation]
    r = Rotation.from_euler("yxz", [data.roll, data.pitch, -data.azimuth], degrees=True)

The pose parsing is adopted from your code with some trial-n-error. Then the point cloud transform between two frames can be obtained by lidar-novatel extrinsics + pose difference + inverse lidar-novatel extrinsics

I noticed that the demo in the dataset website have pretty good alignment, so could you please give an example code for aligning point cloud from two different frames? Thank you!

closed time in 17 days

cmpute

issue commentmpitropov/cadc_devkit

Unable to align the point clouds between two frames

It turned out that there's some bugs in my transformation chaining, and the function I posted above shoud be

def parse_pose_from_inspvax(data: INSPVAX):
   x, y, *_ = utm.from_latlon(data.latitude, data.longitude)
   t = [x, y, data.altitude + data.undulation]
   r = Rotation.from_euler("yxz", [data.roll, data.pitch, -data.azimuth], degrees=True)

Now I can get pretty accurate alignment between point clouds and bounding boxes. Here's my custom loader for the dataset in case anyone is interested: https://github.com/cmpute/d3d/blob/master/d3d/dataset/cadc/loader.py

cmpute

comment created time in 17 days

startedchrdavis/SmartRename

started time in 18 days

startedmarceloprates/prettymaps

started time in 18 days

startedgogojjh/M-LOAM

started time in 18 days

startedPythonOT/POT

started time in 18 days

starteduber/causalml

started time in 18 days

startedkayzhu/LSHash

started time in 18 days

starteducbrise/actnn

started time in 18 days

startedKaiyuYue/torchshard

started time in 18 days

starteddblalock/bolt

started time in 18 days

startedgaopengcuhk/SMCA-DETR

started time in 18 days

startedmicrosoft/FLAML

started time in 18 days

startedhanickadot/compile-time-regular-expressions

started time in 18 days

push eventcmpute/d3d

Jacob Zhong

commit sha 105a1fd96b9973404ac56d4575f5fd84ab7b9740

Various bug fix

view details

push time in 22 days

issue commentmpitropov/cadc_devkit

Unable to align the point clouds between two frames

Thanks for the detailed instructions! I will try again these days

cmpute

comment created time in 24 days

issue commentmpitropov/cadc_devkit

Unable to align the point clouds between two frames

Thanks for the information! Could you elaborate on what's the content of T_ENU_GPSIMU?

cmpute

comment created time in 25 days