meidachen / STPLS3D

🔥 Synthetic and real-world 2d/3d dataset for semantic and instance segmentation (BMVC 2022 Oral)
245 stars 20 forks source link

instance label 0 mean #22

Closed Wangcongyu620 closed 2 years ago

Wangcongyu620 commented 2 years ago

Hi, @meidachen Thank you for sharing a new interesting outdoor dataset for instance segmentation first. I have a little problem that I haven't understood for a long time. For HAIS instance result , if the instance label equal to 0,What does it mean? I found that the instance label 0 appeared in different blocks. I'm confused. Look forward to your reply!

meidachen commented 2 years ago

Hi @Wangcongyu620 ,

Thanks for your interest in our dataset. Instance labeled 0 should be ground I think, as can be seen here:

https://github.com/meidachen/STPLS3D/blob/65917491c6a507b97c1d1ed60dcffd418524e3d8/HAIS/util/utils_3d.py#L64

label 0 is ignored during evaluation.

Wangcongyu620 commented 2 years ago

Thank you for your reply. I understand you,bur I think Instance label 0 should not represent ground.Because its semantic label is not equal to 0 . Now I'm testing my own data. test Semantic label equal to 4 means high vegetation, but the instance label 0,This made my result very bad. When I remove the dot labeled 0, the result is acceptable. So I want to know what the instance label 0 means. Show only the overall results of vegetation 3d2db9ad886cfb559a47779ebb531cb Result of removing tag 0 5a1b6cc0d68f8f714367ce57b715429 It is obvious that all green represents the 0 label,In this code, I can't find what the prediction of 0 means. Sorry to disturb you again, looking forward to your reply!

Lizhinwafu commented 2 years ago

What software did you use to label it? I use cloudcompare.

Wangcongyu620 commented 2 years ago

@Lizhinwafu I also use cloudcompare , just for visualization.

Lizhinwafu commented 2 years ago

I have apple tree point cloud dataset with two semantic classes trunk, branch. One line of my annotation data is as follows. 0.013332, 0.111240, 1.698000,112.000000,109.000000,102.000000, 0.000000, 0.000000

There are 8 columns in total:x,y,z,R,G,B, semantic label(trunk is 0/branch is 1), instance label(trunk is 0, branch1 is 1,branch2 is 2,...).

Wangcongyu620 commented 2 years ago

My dataset is similar. I understand you!So?

Lizhinwafu commented 2 years ago

I'm not sure if this is right,do you know?

Lizhinwafu commented 2 years ago

Can you add QQ to discuss it? 2479929042@qq.com

Wangcongyu620 commented 2 years ago

OK. add you

meidachen commented 2 years ago

@Wangcongyu620 ,

I'm not so sure about what's going on with your data. But one thing I can think of is that instance and semantic labels are merged during evaluation (i.e., semantic*1000 + instance) this is what I was trying to refer to, but not "instance 0". If you haven't thought about this, this may give you a hint, hopefully. Also, I saw in your result, that it seems like the ground was miss segmented as semantic 4 (high veg), and a lot of these grounds were then labeled as instance 0. Not sure, if this is related to ignoring some semantic labels during training.

Sorry, I can't give you a direct answer but only try to help based on my limited knowledge.

Wangcongyu620 commented 2 years ago

I understand. Thank you for your patient answer. Thank you

Greatest-Marx commented 1 year ago

Hi, @meidachen Thank you for sharing a new interesting outdoor dataset for instance segmentation first. I have a little problem that I haven't understood for a long time. For HAIS instance result , if the instance label equal to 0,What does it mean? I found that the instance label 0 appeared in different blocks. I'm confused. Look forward to your reply! Hello! May I ask if you have solved this problem? I am also making my own dataset and am testing it with HAIS, would appreciate if you could get back to me!

Greatest-Marx commented 1 year ago

Can you add QQ to discuss it? 2479929042@qq.com

Hello! I am a second year student at China University of Geosciences, I am also training HAIS with my own dataset, may I add your QQ?