Question about miou

[lsylsylsylsyl]

Thanks for your great work!

I wonder why I have modified the config file as follows, but the miou indicator still does not improve. So could you tell me where else I should modify and how to modify it?

dict(type='LoadOccGTFromFileCVPR2023', scale_1_2=True, scale_1_4=True, scale_1_8=True, ignore_invisible=True),

Thank you very much.

[lzzzzzm]

If you want to reproduce the miou metric of our work, you can change the config setting in stcocc_r50_704x256_16f_occ3d_36e.py
dict(type='LoadOccGTFromFileCVPR2023', scale_1_2=True, scale_1_4=True, scale_1_8=True, ignore_invisible=True),
and retrain the model.

[Rayn-Wu]

Thanks for your great work!

I wonder why I have modified the config file as follows, but the miou indicator still does not improve. So could you tell me where else I should modify and how to modify it?

dict(type='LoadOccGTFromFileCVPR2023', scale_1_2=True, scale_1_4=True, scale_1_8=True, ignore_invisible=True),

Thank you very much.

Hi, I tried to train the model with camera mask too.
I think your need to set self.load_mask=True to enable camera mask.

STCOcc/mmdet3d/datasets/pipelines/loading.py

Lines 52 to 58 in 90c7028

	semantics = occ_labels['semantics']
	if self.load_mask:
	voxel_mask = occ_labels['mask_camera']
	results['voxel_mask_camera'] = voxel_mask.astype(bool)
	if self.ignore_invisible:
	semantics[voxel_mask==0] = 255
	results['voxel_semantics'] = semantics

[lsylsylsylsyl]

After modifying stcocc_r50_704x256_16f_occ3d_36e.py and loading.py according to what you said, should I modify the downsample_label in generate_ms_cocc. py again to generate mutil-scale mask? If so, how should I modify it?

[Rayn-Wu]

After modifying stcocc_r50_704x256_16f_occ3d_36e.py and loading.py according to what you said, should I modify the downsample_label in generate_ms_cocc. py again to generate mutil-scale mask? If so, how should I modify it?

You can refer to this: #8 (comment)

[muchen1021]

After modifying stcocc_r50_704x256_16f_occ3d_36e.py and loading.py according to what you said, should I modify the downsample_label in generate_ms_cocc. py again to generate mutil-scale mask? If so, how should I modify it?

You can refer to this: #8 (comment)

Hello, could you please let me know how I should modify it?
Thank you very much!