Fail to reproduce the result on paper

[lastbasket]

Hi, I ran the code on the Hamlyn and Endomapper dataset, and the evaluation failed to save in the outputs folder.

Setting for the simulation:

%YAML:1.0

Camera model

Camera.model: "PinHole"

Camera calibration and distortion parameters

Camera.fx: 472.64955100886374
Camera.fy: 472.64955100886374
Camera.cx: 479.5
Camera.cy: 359.5

Camera.radiansPerPixel: 0.002

Masking.filterFile: "./data/simulation/filters.txt"

Visualization.cameraSize: 0.5

System.autoplay: 1

Set the below variables to valid paths to save outputs

MapVisualizer.save_path: "outputs/map"
ImageVisualizer.save_path: "outputs/img"
Evaluation.save_path: "outputs/eval"

MapVisualizer.left_view: !!opencv-matrix
rows: 4
cols: 4
dt: f
data: [ 0.724267, 0.688597, -0.0356722, 1.79774,
0.310373, -0.27938, 0.908634, -4.77672,
0.615716, -0.669165, -0.416068, -6.55803,
0.0, 0.0, 0.0, 1.0 ]

MapVisualizer.right_view: !!opencv-matrix
rows: 4
cols: 4
dt: f
data: [ 0.202292, 0.208433, -0.956888, 5.21818,
-0.0225534, -0.975838, -0.217329, 0.277943,
-0.979066, 0.0655449, -0.192703, -5.8181,
0.0, 0.0, 0.0, 1.0 ]

[lrl-hub]

嗨，我在 Hamlyn 和 Endomapper 数据集上运行了代码，但评估无法保存在 outputs 文件夹中。

模拟设置：

%YAML：1.0

相机型号

Camera.model： “针孔”

摄像机校准和失真参数

Camera.fx：472.64955100886374 Camera.fy：472.64955100886374 Camera.cx：479.5 Camera.cy：359.5

Camera.radiansPerPixel：0.002

Masking.filterFile： “./data/simulation/filters.txt”

Visualization.cameraSize：0.5

System.autoplay：1

将以下变量设置为有效路径以保存输出

MapVisualizer.save_path： “outputs/map” ImageVisualizer.save_path： “outputs/img” Evaluation.save_path： “outputs/eval”

MapVisualizer.left_view： ！！OpenCV 矩阵行： 4 列： 4 dt： f 数据： [ 0.724267， 0.688597， -0.0356722， 1.79774， 0.310373， -0.27938， 0.908634， -4.77672， 0.615716， -0.669165， -0.416068， -6.55803， 0.0， 0.0， 0.0， 1.0 ]

MapVisualizer.right_view： ！！OpenCV 矩阵行： 4 列： 4 dt： f 数据： [ 0.202292， 0.208433， -0.956888， 5.21818， -0.0225534， -0.975838， -0.217329， 0.277943， -0.979066， 0.0655449， -0.192703， -5.8181， 0.0， 0.0， 0.0， 1.0 ]

Hello, I've also been researching this paper recently. I'd like to ask if you've resolved the aforementioned issues?
I've encountered two problems:
Dataset issue:
I can't get the Endomapper dataset, but I can access the Hamlyn dataset provided by the authors. However, there isn't much introduction in the videos, so when executing the following code, I'm not sure which video corresponds to which sequence.
复制
./build/bin/hamlyn --dataset_path <video_path>
--settings_path .data/hamlyn_/settings.yaml
--starting_frame <starting_frame>
--end_frame <last_frame>
Output results issue:
I modified the data/hamlyn_01/settings.yaml as follows:
yaml
复制
%YAML:1.0

Camera model

Camera.model: "PinHole"

Camera calibration and distortion parameters

Camera.fx: 766.380279
Camera.fy: 766.380279
Camera.cx: 304.8638076782227
Camera.cy: 258.3343734741211

Camera.radiansPerPixel: 0.002

For stereo evaluation

Stereo.bf: 3886.37

Masking.filterFile: "./data/hamlyn_01/filters.txt"

Visualization.cameraSize: 0.2

System.autoplay: 1

Set the below variables to valid paths to save outputs

MapVisualizer.save_path: "/home/lrl/bag/outputs/MapVisualizer/"
ImageVisualizer.save_path: "/home/lrl/bag/outputs/ImageVisualizer/"
Evaluation.save_path: "/home/lrl/bag/outputs/Evaluation/"

MapVisualizer.left_view: !!opencv-matrix
rows: 4
cols: 4
dt: f
data: [ 0.951395, 0.206095, 0.22885, 1.33195,
-0.0010889, -0.740829, 0.671693, -1.64768,
0.307971, -0.639294, -0.704597, -4.53994,
0.0, 0.0, 0.0, 1.0 ]

MapVisualizer.right_view: !!opencv-matrix
rows: 4
cols: 4
dt: f
data: [ 0.890705, 0.062171, -0.450309, 1.34814,
0.0263112, -0.995994, -0.0854666, -0.24229,
-0.453819, 0.0642774, -0.888773, -3.60393,
0.0, 0.0, 0.0, 1.0 ]
I created the corresponding three folders to save the results, but only the MapVisualizer folder contains outputs after the program finishes running.
I hope to receive your assistance with these two issues. Thank you.

[lastbasket]

@lrl-hub Hi, I think the code for saving the result is not working (not completed), so I have modified them for saving the trajectory. For the EndoMapper and Hamlyn datasets, I am also confused about the sequence.

[lrl-hub]

OK, thank you for your reply. It's my fault for being slow-witted. So I'd like to ask whether the results can be saved for the code you've modified. For the data input into this program by the author, there is only one video and setting, and there is no input of the ground truth. Can it output the evaluation results? Ruiliang&nbsp;Liu ***@***.*** &nbsp;

…

------------------&nbsp;原始邮件&nbsp;------------------ 发件人: "endomapper/NR-SLAM" ***@***.***&gt;; 发送时间:&nbsp;2025年3月19日(星期三) 下午5:38 ***@***.***&gt;; 抄送:&nbsp;"Ruiliang ***@***.******@***.***&gt;; 主题:&nbsp;Re: [endomapper/NR-SLAM] Fail to reproduce the result on paper (Issue #5) @lrl-hub Hi, I think the code for saving the result is not working (not completed), so I have modified them for saving the trajectory. For the EndoMapper and Hamlyn datasets, I am also confused about the sequence. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***&gt; lastbasket left a comment (endomapper/NR-SLAM#5) @lrl-hub Hi, I think the code for saving the result is not working (not completed), so I have modified them for saving the trajectory. For the EndoMapper and Hamlyn datasets, I am also confused about the sequence. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***&gt;

[lastbasket]

@lrl-hub Certainly, the evaluation result can not be produced without the ground truth. I think the only way is to add the ground truth in the loader and do the evaluation. My pull request only provides the estimation trajectory result.

[lrl-hub]

当然，如果没有基本事实，就无法产生评估结果。我认为唯一的方法是在 loader 中添加 ground truth 并进行评估。我的 pull request 只提供了 estimation trajection 结果。
Thank you, I understand now.

[yahiaboter]

I saw in your added code to save the trajectory and I noticed that you try to use the TUM dataset as ground throut do you have any idea about how to use ( do you use the evolution of the L refrence over time )? do you know how to synchronise the estimated trajecotry with the ground throut (time is negatibe event when we use the 100 ms of time shift ? please if you have any idea dont hesitate to share it. thank you

[lrl-hub]

I saw in your added code to save the trajectory and I noticed that you try to use the TUM dataset as ground throut do you have any idea about how to use ( do you use the evolution of the L refrence over time )? do you know how to synchronise the estimated trajecotry with the ground throut (time is negatibe event when we use the 100 ms of time shift ? please if you have any idea dont hesitate to share it. thank you

While studying the author's code, I found that this system may not require the input of ground truth. The ground truth should be generated in utilities/frame_evaluator.cc. You can try to modify the code in SLAM/system.cc to call this evaluation function. I hope this helps you.