longkun

Results 16 comments of longkun

A question about Formula(6) in the paper.

Thanks for your reply! It seems that h_i_s and h_i_t just mean SA node features, and they are obtained by concatenating all node features together since it can be seen...

manual installation fails

I have met the same issue. pytorch 1.9.1 cuda 11.1

RuntimeError: Expected object of type torch.FloatTensor but found type torch.cuda.FloatTensor for argument #4 'other'

Hi, I met the same problem with you，have you resolved it?

EOFError: Ran out of input

@TinyGrass I have resolved it .The reason for this problem is that the file MT-LSTM.path is empty, you need to download this file again. You can excute the command line...

Supervised contrastive loss saturation from the beginning on custom data

Hi，I encountered the same problem, did you solve this problem? @zaid478

Supervised contrastive loss saturation from the beginning on custom data

When training, please turn on the argument"--warm" or set a larger learning rate（--learning_rate 0.5）, it seems to solve this problem, at least it works for me

questions about effect of number of positives

You seem to be talking about self-supervised contrastive learning, right? Just like SimCLR

Unable to reproduce the good results of shapenet to scannet without SPST

Hi, I have uploaded some run.log files, hope this helps.

Run Problem

Hi, before starting training, you need to run the *.py files under the data file for data processing.

Run Problem

@KouGZ The ".DS_Store" files in the dataset are leftover from the macOS system, you need to clear these files.