search

ctu-vras / depth_correction

Self-Supervised Depth Correction of Lidar Measurements from Map Consistency Loss
BSD 3-Clause "New" or "Revised" License
10 stars 0 forks source link

Question about fee_corridor data set. #3

Open xiaojxkevin opened 10 months ago

xiaojxkevin commented 10 months ago

Dear authors,

After downloading the fee_corridor data set, I notice that there are two folders of point clouds in sequences. One is named as ouster_points and the other is static_outster_points, the former on consists of around 100 frames while the later one consists of around 30 frames. I'm wondering what are the differences between these two folders. Do they come from different environments, or simply the way they sampled are different. And if the former one is sampled when the vehicle is moving, there might be the distortions of the points, then how could we deal with such case?

Thanks!

RuslanAgishev commented 10 months ago

Hi,

The "static" poses (where the robot did not move) are provided in static_poses.csv and static_ouster_poses. Additionally, the ouster_points and corresponding poses.csv files contain point cloud and localization waypoints recorded in the same environment and distant approximately 1m apart (without deskewing). The former data recording method was used to mitigate an error caused by the robot's movement. Thanks for the question.

xiaojxkevin commented 10 months ago

Since there are distortions in ouster_points, then what are these clouds used for? Are they used in the depth-correction process?

Thanks!

RuslanAgishev commented 10 months ago

We used the static clouds for the depth correction models training. The ouster_points clouds could be used for SLAM evaluation.

xiaojxkevin commented 10 months ago

Hi! When I try to reproduce your SLAM evaluation (run slam_eval.launch), I just simply use the points from the bag file (fee_corridor/bags/seq1.bag). So I am wondering if the points in the bag file are equivalent to points provided in ouster_points

RuslanAgishev commented 10 months ago

The provided bag-files contain all the point clouds recorded at 10hz. The points folder contains a sub-set of them (we save the clouds every 1m traveled by the robot).

xiaojxkevin commented 10 months ago

Hi, thanks for the reply! In fact, when I try to reproduce your SLAM evaluation results, I notice that there are a slight differences with the results presented in the paper. Thus, I am wondering if you could sent me a copy of the pretrained model of yours.

RuslanAgishev commented 10 months ago

Good point! We will upload the pre-trained model weights. Thanks

xiaojxkevin commented 10 months ago

Thanks for your replies.

xiaojxkevin commented 9 months ago

Hi, sorry for bothering, I also notice something quite strange. In the given fee_corridor data set, all the .npz files represent PointCloud data, and their formats should be (n, 4), where n means the number of points, and 4 is number of values in a certain row. I can tell that the first 3 values are x,y,z, and I am wondering what is the meaning of the 4th value, is it something like intensity?