metaslam / gpr_competition Goto Github PK

I am willing to take part in the competition. Could you tell me when will datasets be available for download？
Thanks！

@maxtomCMU Hi,the work is excellent!But I wan to know the package gpr.evaluation how to get PR curve,the code only can get the Topk-recall.Thanks!

I've noticed that on competion site (https://sites.google.com/andrew.cmu.edu/gpr-competition/datasets) you've described "Pitts Helicopter Dataset" as a dataset for second task, but in repository's readme you've described another dataset "CMU Cross-Domain Dataset". More of that, in dataset_description.md there is again description of "Pitts Helicopter Dataset".

Could you clarify what information is relevant? Thank you!

When will datasets be available for download？

Thank for this dataset.

We have a pipeline and are able to achieve good results on the validation set. However, our method does not result in a descriptor for every point cloud instead for a given point pair of point clouds we directly return the distance between them in the latent space.

As I understand the current submission format does not support this, can you please provide a few instructions to upload results of our new method

There is no gt_matches.csv for Train subset of UAV dataset for round 2:

$ ls Train/
query.csv  query_images  reference.csv  reference_images

$ ls Val/
gt_matches.csv  query.csv  query_images  reference.csv  reference_images

Can you provide this file? Thank you!

Dear authors, I still can not download the dataset, please provide downloadable resources, thank you very much.

The links for the dataset download are not working.

Thanks for your excellent dataset.
I want to do some experiments on your excellent dataset for Place Recognition but I have a question: the submap in the dataset doesn't have absolute location because only relative poses of submaps are given, long-term information of multiple trajectories can't be recognized. So I ask that if you can release the GPS or absolute location of every submaps in every track? Thank you!

Hi,

I'm currently trying to download the dataset in Australia - I cannot register an account for Baidu or navigate the interface, and the dropbox link still appears to be broken. Is there a timeline on when this issue will be fixed?

Thank you,

Joshua

Somehow I can't download the training dataset for UAV Round 2. I use this dropbox link. I could click the download, it progressed, but at the end of the download, it was suddenly failing. I tried multiple times using Firefox and Chrome. I have no problem with the validation and testing dataset.

I am unable to extract the files TRAIN.tar.gz.00 , TRAIN.tar.gz.01 , TRAIN.tar.gz.02 for the 3D-3D LiDAR scan dataset given in this link, can you please help me with this.

Thank You

Why does "true_neighbors" use indice of queries_feature instead of reference_feature? Isn't "retrieved_indices" from "reference_feature"?

I try to use the pose in npy file to visualize database and query clouds. In round 1, the paired point clouds are successfully displayed in picture above. However, in round 2, the point clouds show a clear deviation in picture below. I guess there is a mismatch between pose and point cloud

Dear authors, please provide downloadable resources, thank you very much.

Thank you for sharing great dataset.
Now I'm trying to recover approximate metric unit translation from consequtive two images using Homography.
For example, if translation btw two images(query 00001.png, query 00004.png) is (10pixel, 30pixel), i want to recover approximate metric unit translation using meter/pixel scale information.

So, I would like to request meter/pixel information for the duration of the flight path if that information exists. (for both aerial / satellite caes.)
Thank you in advance.