yanyan-li / Structure-SLAM-PointLine

This is a basic point-line SLAM system based on ORBSLAM2.
373 stars 71 forks source link

关于ICL-NUIM数据集使用的问题 #4

Open seanleecn opened 3 years ago

seanleecn commented 3 years ago

您好,感谢您的工作! 我在使用中发现,Structure-SLAM.cc文件中用string strFile = string(argv[3]) + "/rgb.txt";方法读取数据集图片,但是在ICL-NUIM数据集中并没有看见rgb.txt文件 请问我应该怎样正确读取ICL数据集的图片,感谢您!

heebinYoo commented 3 years ago

did you solve it? how can I find rgb.txt? in my case originally rgb.txt doesn't exist in dataset

seanleecn commented 3 years ago

did you solve it? how can I find rgb.txt? in my case originally rgb.txt doesn't exist in dataset

i found that the timestamp in groundtruth file is in the format below, so i made a rgb.txt in this format.

0.1 rgb/0.png
1   rgb/1.png
2   rgb/2.png
3   rgb/3.png
4   rgb/4.png
5   rgb/5.png
6   rgb/6.png

But I'm not sure if it is correct, because when i testing this system, it just breakdown after importing maybe 10 pictures~

I will reopen this issue, maybe others can give more advice~

Dkaka commented 3 years ago

It looks like the LoadImages() function is very similar to the mono_tum.cc of ORBSLAM2. https://github.com/raulmur/ORB_SLAM2/blob/f2e6f51cdc8d067655d90a78c06261378e07e8f3/Examples/Monocular/mono_tum.cc#L128 So the rgb.txt should follow the same syntax of rgb.txt of TUM Dataset. https://vision.in.tum.de/data/datasets/rgbd-dataset/download

tested with fr1/xyz dataset and it works fine.

an example of rgb.txt from fr1/xyz:

# color images
# file: 'rgbd_dataset_freiburg1_xyz.bag'
# timestamp filename
1305031102.175304 rgb/1305031102.175304.png
1305031102.211214 rgb/1305031102.211214.png
1305031102.243211 rgb/1305031102.243211.png
1305031102.275326 rgb/1305031102.275326.png
1305031102.311267 rgb/1305031102.311267.png
1305031102.343233 rgb/1305031102.343233.png

And the timestamps should be in seconds with the interval matching the fps. 30fps for TUM and ICL-NUIM.

pansyl commented 3 years ago

您好,我和你有一样的问题,运行TUM数据集效果很好,然而运行ILC-NUIM数据集时,并没有能够正确读取数据集的图片,我也很想知道怎么使用ICL数据集,请问您有办法了吗?

yanyan-li commented 3 years ago

您好,我和你有一样的问题,运行TUM数据集效果很好,然而运行ILC-NUIM数据集时,并没有能够正确读取数据集的图片,我也很想知道怎么使用ICL数据集,请问您有办法了吗?

For the ICL dataset, their association.txt files may need to be modified. Make sure to provide rgb and depth images for the correct entries in the file Structure-SLAM.cc.

pansyl commented 3 years ago

您好,我和你有一样的问题,运行TUM数据集效果很好,然而运行ILC-NUIM数据集时,并没有能够正确读取数据集的图片,我也很想知道怎么使用ICL数据集,请问您有办法了吗?

For the ICL dataset, their association.txt files may need to be modified. Make sure to provide rgb and depth images for the correct entries in the file Structure-SLAM.cc.

it looks like this in association.txt 0 depth/0.png 0 rgb/0.png 1 depth/1.png 1 rgb/1.png 2 depth/2.png 2 rgb/2.png 3 depth/3.png 3 rgb/3.png 4 depth/4.png 4 rgb/4.png 5 depth/5.png 5 rgb/5.png 6 depth/6.png 6 rgb/6.png 7 depth/7.png 7 rgb/7.png 8 depth/8.png 8 rgb/8.png 9 depth/9.png 9 rgb/9.png

Do you mean to convert it to the same format as TUM?

1790426751 commented 1 year ago

Hello, may I ask if this system can run with only RGB images as input without depth information?

wz0515 commented 6 months ago

请问以下,TUM数据集我跑几秒中,就读取了几十帧图像程序就终止了,请问如何解决

brz-hub commented 4 months ago

這個是按照orbslam2的mono_tum.cc改的,所以如果你想跑kitti數據集或者euroc數據集,直接從orbslam2的歷程中仿照這來就行。我也是跑着跑着就斷了,不知道爲什麼