search

qianguih / RSNet

This is the official implementation of RSNet.
Other
118 stars 42 forks source link

Testing results don't match the original room data. #2

Closed Hao-FANG-92 closed 6 years ago

Hao-FANG-92 commented 6 years ago

Hi, qian gui

Thanks for the share of this excellent work and it's really nice! I have some questions on the testing results on Area5.

Actually, the visualization results in folder ./results, i.e. Area_5_conferenceRoom_1_gt seems like a different room with the RAW_DATA/Area5/conferenceRoom_1. It's the same case for the other rooms.

Do you have idea about that?

qianguih commented 6 years ago

thanks for pointing this out. Yes, there is a mismatch in the file name. our internal version works fine. I am hunting potential bugs to fix this. will be fixed in next commit.

Hao-FANG-92 commented 6 years ago

Hi, qian gui

I think the bug is in line 54 and 55 of file load_data.py, where the return list of h5 files are not odered. Change it as: train_files = sorted(glob.glob(TRAIN_DIR + '.h5')) test_files = sorted(glob.glob(TEST_DIR + '.h5')) and the problem is fixed.

Thanks for your time.

tonysy commented 6 years ago

@Hao-FANG-92 Hi, could you share the solution to visualize the .obj file ? I'm new to 3D vision and not familiar with this. Thanks a lot.

iSarmad commented 6 years ago

Hi these obj files cannot be visualized using Obj file viewers available, there is something wrong with the files generated by the code. Can you tell how you visualize these files ? thanks

longmalongma commented 5 years ago

Hi, qian gui

Thanks for the share of this excellent work and it's really nice! I have some questions on the testing results on Area5.

Actually, the visualization results in folder ./results, i.e. Area_5_conferenceRoom_1_gt seems like a different room with the RAW_DATA/Area5/conferenceRoom_1. It's the same case for the other rooms.

Do you have idea about that?

Hi,when I try to run your code,I find I can not compile your slice_pool_layer and slice_unpool_layer,can you send me your files have compiled files?my email is that:1442342449@qq.com.looking forward your reply

longmalongma commented 5 years ago

.obj fil

meshlab is ok.

MCemGunes commented 2 years ago

Hello all, sorted() Method worked on my local machine but I still got mixed results on the HPC server. I don't know the exact problem but it might be multiprocessing or a similar thing. Not sure but, it is not that important.

I want to share my solution in case of anyone faces a similar problem.

We can use the number of the ply_dataall.h5 files to sort the files directly. * might be 1,2,...,29,215,... basically any integer.

Simply, we normalize the path which means we transform it into a string (it is optional). Then we split the text using the "_" characters and choose the last part (ie. 223.h5 ). Finally, we split the chunk we have into two using the "." character and pick the first part. I use my custom dataset so the naming convention might vary. The main idea is sorted() function alone is error-prone however, you can be sure you use the right order by adding a key that sorts integers inside the name directly. If you wish you can implement the key inside the first sorted() methods.

The solution adapted from this post.

train_files = sorted(train_files, key=lambda x: int(  (os.path.normpath(x).split("_"))[-1].split(".")[0]  ))
test_files = sorted(test_files, key=lambda x: int(  (os.path.normpath(x).split("_"))[-1].split(".")[0]  ))