jokieleung / transpcc

the Pytorch implementation for our ICMR 2022 paper "TransPCC: Towards Deep Point Cloud Compression via Transformers"

Hi 👋, I'm Zujie Liang

Welcome! You are the visitor!

• 😄 Hi there, this is Zujie Liang (or "Jokie Leung" in Cantonese). I am a master student at Sun Yat-Sen University (SYSU).
• 🔭 My research interest includes Natural Language Processing, Machine Learning, and Computer Vision.
• 👯 I am open to collaboration on research, feel free to contact me via E-mail!
• ⚡ Please check my homepage for my CV and latest update!

transpcc's Issues

Decoder - Inputs

Hello,

In your code for TransitionUp(

TransPCC/depoco/architectures/network_blocks.py

Line 485 in 1c681ab

points = self.transition_ups[i](xyz, points, xyz_and_feats[- i - 2][0], xyz_and_feats[- i - 2][1])

) you use the points from previous encoder stages for interpolation. In your paper I didn't see any mentions of it and for the results did you also calculate the contribution of these points from encoder to the overall compressed representation size ?

Could you please clarify these questions ?

Thanks

Misleading code/results?

Hi,
I'm trying to figure how you are doing pointcloud upsampling on the decoder side. I see that you are feeding the downsampled pointcloud from the encoder stage to TransitionUp in the decoder

[TransPCC/depoco/architectures/network_blocks.py]
points = self.transition_ups[i](xyz, points, xyz_and_feats[- i - 2][0], xyz_and_feats[- i - 2][1])

I do not understand how you are doing the upsampling and compression result analysis presented in your paper while working with the input pointclouds directly. Could you please clarify If my understanding is wrong and where I can find the "Right" implementation?

Wrong Encoder input

Your encoder input is coordinate information of point cloud , ignore the feature of point cloud.
Or I misunderstood.