There are many small inaccuracies in the segmentation results on my dataset. Could you please provide some suggestions?

JonasSchult / Mask3D

Mask3D predicts accurate 3D semantic instances achieving state-of-the-art on ScanNet, ScanNet200, S3DIS and STPLS3D.

MIT License

525 stars 103 forks source link

There are many small inaccuracies in the segmentation results on my dataset. Could you please provide some suggestions? #157

Open clawCa opened 5 months ago

clawCa commented 5 months ago

Thank you for open sourcing your code! I trained the model on my own dataset. In my segmentation scenario, I have a class representing over 30 potted plants. the average precision (AP) value for this class is an impressive 0.9. However, upon reviewing the segmentation results, I noticed the presence of many small instances resembling speckles on each potted plant. Could you provide some tuning advice?

narges-tk commented 4 months ago

Thank you for open sourcing your code! I trained the model on my own dataset. In my segmentation scenario, I have a class representing over 30 potted plants. the average precision (AP) value for this class is an impressive 0.9. However, upon reviewing the segmentation results, I noticed the presence of many small instances resembling speckles on each potted plant. Could you provide some tuning advice?

Hi! May I please ask you to share the versions of the modules that are working for you? I am getting the error especially for hydra. It seems that the reported versions in environments no longer work. Thanks in advance for your help!

clawCa commented 4 months ago

Hi! May I please ask you to share the versions of the modules that are working for you? I am getting the error especially for hydra. It seems that the reported versions in environments no longer work. Thanks in advance for your help!

hydra-core == 1.0.5, maybe you can try offline mode. environment_mask3d.zip

narges-tk commented 4 months ago

Hi! May I please ask you to share the versions of the modules that are working for you? I am getting the error especially for hydra. It seems that the reported versions in environments no longer work. Thanks in advance for your help!

hydra-core == 1.0.5, maybe you can try offline mode. environment_mask3d.zip

Many thanks for your great help!

narges-tk commented 4 months ago

Thank you for open sourcing your code! I trained the model on my own dataset. In my segmentation scenario, I have a class representing over 30 potted plants. the average precision (AP) value for this class is an impressive 0.9. However, upon reviewing the segmentation results, I noticed the presence of many small instances resembling speckles on each potted plant. Could you provide some tuning advice?

Hi! May I please ask to guide me through how to visualize segmentation results? In fact, the following figure is the content of my visualization folder. However, when I open the .bin files of segmentation in cloud compare it is a huge file for even a very small subset and can not be opened. Thanks in advance for your help.

clawCa commented 4 months ago

Thank you for open sourcing your code! I trained the model on my own dataset. In my segmentation scenario, I have a class representing over 30 potted plants. the average precision (AP) value for this class is an impressive 0.9. However, upon reviewing the segmentation results, I noticed the presence of many small instances resembling speckles on each potted plant. Could you provide some tuning advice?

Hi! May I please ask to guide me through how to visualize segmentation results? In fact, the following figure is the content of my visualization folder. However, when I open the .bin files of segmentation in cloud compare it is a huge file for even a very small subset and can not be opened. Thanks in advance for your help.

Based on the generated bin file names, find the code part that stores the point cloud. You need to add the code to save the point cloud and its labels. In another issue, the author replied about the methods for visualization.

narges-tk commented 4 months ago

Thank you for open sourcing your code! I trained the model on my own dataset. In my segmentation scenario, I have a class representing over 30 potted plants. the average precision (AP) value for this class is an impressive 0.9. However, upon reviewing the segmentation results, I noticed the presence of many small instances resembling speckles on each potted plant. Could you provide some tuning advice?

Hi! May I please ask to guide me through how to visualize segmentation results? In fact, the following figure is the content of my visualization folder. However, when I open the .bin files of segmentation in cloud compare it is a huge file for even a very small subset and can not be opened. Thanks in advance for your help.

Based on the generated bin file names, find the code part that stores the point cloud. You need to add the code to save the point cloud and its labels. In another issue, the author replied about the methods for visualization.

Thanks a lot for your prompt help!

bh-cai commented 3 months ago

Thank you for your issues,may I know how your dataset is made, and how to use it?

bh-cai commented 3 months ago

Thank you for open sourcing your code! I trained the model on my own dataset. In my segmentation scenario, I have a class representing over 30 potted plants. the average precision (AP) value for this class is an impressive 0.9. However, upon reviewing the segmentation results, I noticed the presence of many small instances resembling speckles on each potted plant. Could you provide some tuning advice?

Thank you for your issues,may I know how your dataset is made, and how to use it?

clawCa commented 3 months ago

Thank you for your issues,may I know how your dataset is made, and how to use it?

You need to modify the pre-processing code provided by the author based on the format of your own dataset.

Lizhinwafu commented 2 months ago

What's your training code and testing code?