search

d62lu / 3DGTN

3DGTN: 3D Dual-Attention GLocal Transformer Network for Point Cloud Classification and Segmentation
MIT License
11 stars 0 forks source link

accuarcy #1

Open Wanglei794 opened 8 months ago

d62lu commented 8 months ago

Hi! Thanks for your interest. This paper is developed based on 3DCTN. Could you please tell me where you think the code is inconsistent with the description in the paper？ Also, could you please show me your log file that has low accuracy?

Wanglei794 commented 8 months ago

sry.i deleted it as soon as I posted the comment.i ignore some details when i looked up the paper. and acc lower than your paper maybe cause i overlook label smoothing.  sry again. nice writing, guy

---Original--- From: @.> Date: Wed, Jan 31, 2024 01:30 AM To: @.>; Cc: @.**@.>; Subject: Re: [d62lu/3DGTN] accuarcy (Issue #1)

Hi! Thanks for your interest. This paper is developed based on 3DCTN. Could you please tell me where you think the code is inconsistent with the description in the paper？ Also, could you please show me your log file that has low accuracy?

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

d62lu commented 8 months ago

No worries. Feel free to discuss if you have any questions 👍 .

Wanglei794 commented 8 months ago

3dgtn_cls.txt my log file, I add the normal feature.got OA 93.30%. But its still the best res in various methods, Because running it on my device(4060Ti),Most of them do not achieve 93% OA

d62lu commented 8 months ago

Great! You can use pre trained model showed in the git, it will make the result better~ 👍

Wanglei794 commented 8 months ago

em...,I am a rookie,Do you mean to use your best_model file? In this case, I'd better give up. Thank you for your answer