hehefan / point-spatio-temporal-convolution Goto Github PK
View Code? Open in Web Editor NEWImplementation of the "PSTNet: Point Spatio-Temporal Convolution on Point Cloud Sequences" paper.
License: MIT License
Implementation of the "PSTNet: Point Spatio-Temporal Convolution on Point Cloud Sequences" paper.
License: MIT License
Great works! I've read it thoroughly and can easily run your code.
One issue is when I directly use the train-msr
and finish train, the final acc is 84 and 83 for two repeat. I think they are with 16 frames. My repeats get lower scoring comparing to the paper.
I think it might because there are some parameter change in the script.
Would you be convenient to check out?
Hi, this is a nice paper. Could you release the codes of (model, dataset, training process) on NTU RGB+D 60 and 120?
Thanks.
Thanks for your awesome work and generous code sharing.
Recently, I have tried to reproduce the results on NTU 60 XSUB benchmark with one 3090 GPU.
However, it seems the training is really slow and it would take more than a week to train 20 epochs following the original setting.
I have tried to enlarge the batch size, and it still quite slow.
So, what is the reasonable training time for NTU 60 dataset?
Is there anything I have missed?
Looking forward to your reply.
Thanks
Hi there,
Thanks for your interesting work. Can you provide the dataset file for the NTU RGBD 60 cross view setting?
Regards,
Jinyang
Hi, @hehefan ,
I've read your papers "PSTNet" and "P4Transformer". Thank you for the awesome work.
I have tried training both of them on an other dataset "BMLrub" from the "Amass" archive.
I transformed the data used the same approach to process and load the datafrom mesh to PC, then used the same approach to process and load the data to train both models.
"P4Transformer" gives me 91-92% accuracy, but "PSTNet" gives only 35-36%.
Could it be the similarity of action classes ('walk in circle', 'normal walk', 'thridmill slow',thridmill normal',thridmill fast', ...), that makes it difficult for the convolution based model to differentiate between them? or must it be something wrong in the modifications I did ?
If there is anything I can add to clear the problem please let me know.
Thank you for your time.
hi, i find "from datasets.msr import MSRAction3D" in train-msr.py. Could you upload the folder ./datasets
I have followed your series of work and have great interest in research. Can you upload the code related to the training of synthia4d dataset? Thank you very much. It will be very helpful to my study. @hehefan
Hi, @hehefan ,
Thanks for releasing the package. Would you also release a basic step-by-step guide for the usage of PSTConv? e.g, data preparation, training steps and prediction steps...
Thanks~
Hi @hehefan,
Thank you for your work and for making the PSTNet code available.
I am trying to reproduce the results on MSR 3d Action Data but running into "PSTConv: Temporal parameter error!".
I had to modify some parameters for conv2a and conv3a layers of the MSR network to make it work, but that gives me ~70% accuracy in 35 epochs training.
Would you have any advice on what would be the correct way to fix the kernel dimensions?
Thanks!
Dear Author,
Hope you are well.
I got an issue related to the pointnet2_utils.py and setup.py files. When I stalled setup.py the file, I got a warning below related to the GCC version.
Then when I was running the highlighted part within the setup.py below, I got an error of "Process finished with exit code 139 (interrupted by signal 11: SIGSEGV)".
Just wondering if you happen to know what's the issue? Does it related to the g++version?
Our PyTorch version is 1.2.0.
Many thanks! @hehefan
Can the author inform the parameters and FLOPs of the article
Hi Hehe,
Can you pleaase share the npy data file for the MSR data set.
It is a bit struggle for me to phrase bin data into npy.
Hope you can help me.
Best,
Xin
Great work on the Point Tubes. I am particularly interested in the 4D semantic segmentation applications.
I was wondering if you tried the PSTNet on benchmark datasets like Semantic Kitti or nuScenes dataset.
These pointcloud sequences are much more sparse than the SYNTHIA dataset.
Thank you
Hi,
Could you please share the values of the different hyperparameters to get the baseline results in the paper for MSRA?
For 24 frames, I am getting an accuracy of 90.24 after 35 epochs:
batch_size: 16
lr: 0.01
lr_gamma: 0.1
lr_warmup_epochs: 10
momentum: 0.9
nsamples: 9
num_points: 2,048
radius: 0.5
Could you confirm that they are the same for PSTNet++?
Thanks a lot!
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.