How to do inference on my own pointcloud?

[zhiwenfan]

Thanks for your sharing your code. I wish to know how to run this code on my own pointcloud set. Since I did not find that script and ./mask-rcnn/inference_room.py need the annotation. However my own dataset does not has annotation.

[woodfrog]

Hi, you could try to adapt the data preprocessing script according to your own data format, and you can simply remove all code related to the annotation processing for running inference only.

[ghost]

@woodfrog Hi, thanks for your work, it looks awesome!
I have a question about custom data and corner detection model.
So, I have my custom point clouds (it's not an apartment) and when I make an inference of the corner detection, I'm getting weird results. I understand that the corner model was train on let say apartment dataset and I'm inferencing something a bit different. And still, I'm convinced that the result must be better than I get.
This are some images on density map and corner prediction. I tried to generate the data with different number of point clouds to be processed [50k, 250k, 1ml]. I'm also curious why density map looks different (less intense) on corner prediction results .

I would appreciate any suggestions.
Thanks.

[woodfrog]

@volodymyrkepsha Hi, your first density map looks all right, and I believe the pretrained Mask-RCNN and corner/edge models should be able to produce okay results for that one. There might be something wrong in your preprocessing steps, and the data fed to the model is not in the required format(i.e., different from the one used for training). Did you try to run the same code on the Beike data?

[ghost]

@woodfrog
Thanks for the answer!
Yes, before running the pipeline on my data, I tried the one you mentioned in the readme. Everything worked fine.
About pre-processing. As you suggested, I removed all the code for annotation processing.
Also, the very interesting thing about the picture with red dots. As I wrote in my prev. post I tried to generate the image data with different number of point clouds, and all the time I got a bit different heat maps (brighter or darker), but the generated image (with red dots) always remains the same, even the red dots position is always the same.
Maybe I made a mistake somewhere, so I'll re-check everything,
Anyway, thanks for the great work! Keep it up !

[xiecookie]

@volodymyrkepsha
Hello！
I am trying to use the project.Could I ask whether you succeed in running the project without annotation？

[ghost]

@xiecookie
Hi. Yes I did.
I just removed all the code for annotations and fixed few things

[y5wang]

@volodymyrkepsha
Is it possible / do you mind posting your preprocessing code without annotation?

[ghost]

@y5wang Hi,
Mostly, I updated inference_room.py and inference_corner.py
I'll provide codeshare link, because here it looks messy, it expires in 24h, so ping me if its gone
https://codeshare.io/G7R071

[y5wang]

Thank you, @volodymyrkepsha. Appreciated. Will see if I can get something going and share with the community.

[shingkim]

@y5wang I also encountered the same problem. Can you share some solutions?thanks

[marcomiglionico94]

@y5wang Hi, Mostly, I updated inference_room.py and inference_corner.py I'll provide codeshare link, because here it looks messy, it expires in 24h, so ping me if its gone https://codeshare.io/G7R071

Hello do you still have that part of the code? If yes can you provide it? Thanks a lot

[y5wang]

@marcomiglionico94 Sorry, it's been a while. Can't find it in my main computer. I will have to search my backups.

[marcomiglionico94]

@y5wang if you can find it would be great. Thanks a lot

[tejaswivg]

@marcomiglionico94 and @shingkim

@y5wang Hi, Mostly, I updated inference_room.py and inference_corner.py I'll provide codeshare link, because here it looks messy, it expires in 24h, so ping me if its gone https://codeshare.io/G7R071

https://gist.github.com/tejaswivg/4f8edebaa4ea06078cb84338a2617609
and
https://gist.github.com/tejaswivg/8d05910dce3ded9b580ca21cc4813a93

I formatted their code from the edit history.

[marcomiglionico94]

@marcomiglionico94 and @shingkim

@y5wang Hi, Mostly, I updated inference_room.py and inference_corner.py I'll provide codeshare link, because here it looks messy, it expires in 24h, so ping me if its gone https://codeshare.io/G7R071

https://gist.github.com/tejaswivg/4f8edebaa4ea06078cb84338a2617609 and https://gist.github.com/tejaswivg/8d05910dce3ded9b580ca21cc4813a93

I formatted their code from the edit history.

Great thanks a lot

[josematez]

@marcomiglionico94 and @shingkim

@y5wang Hi, Mostly, I updated inference_room.py and inference_corner.py I'll provide codeshare link, because here it looks messy, it expires in 24h, so ping me if its gone https://codeshare.io/G7R071

https://gist.github.com/tejaswivg/4f8edebaa4ea06078cb84338a2617609 and https://gist.github.com/tejaswivg/8d05910dce3ded9b580ca21cc4813a93

I formatted their code from the edit history.

Thanks! and you would have the data_writer.py code without annotations?

[kapilrs2]

I am getting errors upon implementing this method. It would be great if someone can assist me. Look forward to connect to you.

[kapilrs2]

@woodfrog Thanks for the answer! Yes, before running the pipeline on my data, I tried the one you mentioned in the readme. Everything worked fine. About pre-processing. As you suggested, I removed all the code for annotation processing. Also, the very interesting thing about the picture with red dots. As I wrote in my prev. post I tried to generate the image data with different number of point clouds, and all the time I got a bit different heat maps (brighter or darker), but the generated image (with red dots) always remains the same, even the red dots position is always the same. Maybe I made a mistake somewhere, so I'll re-check everything, Anyway, thanks for the great work! Keep it up !

@volodymyrkepsha I need some help in implementing this method. Can you let me know whats the bets way to connect?