poor result on my own dataset

[takeshie]

did anyone met the problem that using custom data with a poor result?I use neus-facto and others like neus-facto-angelo get a very strange result

I use this parameter settings : ns-train neus-facto --pipeline.model.sdf-field.inside-outside False --vis viewer --experiment-name gold_ring_neuralangelo1 --data /data/yth/compare_work/sdfstudio-master/data_ours/gold_ring --viewer.websocket-port 7001 --trainer.steps-per-save 2000 --pipeline.model.sdf-field.bias 0.1 --vis viewer

the result and the datalink like as follow: 链接：https://pan.baidu.com/s/1DlP2jWEZegxsjyawy5gChQ?pwd=zvan 提取码：zvan the result and image like as follow: https://s2.loli.net/2023/06/29/cRjpIVi7mnrMXkE.png https://s2.loli.net/2023/06/29/IloOnJVqxefPEcT.png

[niujinshuchong]

@takeshie Hi, your data has white background. Maybe you could change the background color to white during training with --pipeline.model.background-color white. Could you share your data in google drive (or others) so that i can download directly?

[takeshie]

@takeshie Hi, your data has white background. Maybe you could change the background color to white during training with --pipeline.model.background-color white. Could you share your data in google drive (or others) so that i can download directly?

thank you for your respond~I will upload that,just wait me 10 min. I have tried with white background it still not change

[takeshie]

@niujinshuchong hi~I have updated the raw dataset under this link: https://drive.google.com/drive/folders/1rIrgbF-0TY2_LI0a8WZZ0HoiYW6_3_PK?usp=sharing

I process the data by"process_nerfstudio_to_sdfstudio.py",the command follow this: python scripts/datasets/process_nerfstudio_to_sdfstudio.py\ --data /data/yth/compare_work/studio_yth/sdfstudio-master/data_ours/gold_ring001/gold_ring001_raw\ --output-dir /data/yth/compare_work/studio_yth/sdfstudio-master/data_ours/gold_ring001/gold_ring001_process\ --scene-type object\ --data-type colmap

[MvWouden]

@niujinshuchong hi~I have updated the raw dataset under this link: https://drive.google.com/drive/folders/1rIrgbF-0TY2_LI0a8WZZ0HoiYW6_3_PK?usp=sharing

I process the data by"process_nerfstudio_to_sdfstudio.py",the command follow this: python scripts/datasets/process_nerfstudio_to_sdfstudio.py --data /data/yth/compare_work/studio_yth/sdfstudio-master/data_ours/gold_ring001/gold_ring001_raw --output-dir /data/yth/compare_work/studio_yth/sdfstudio-master/data_ours/gold_ring001/gold_ring001_process --scene-type object --data-type colmap

Have you also tried BakedSDF? I found it gives me better results for object-centric data. I would also try the parameter --pipeline.model.sdf-field.inside-outside False

[takeshie]

@niujinshuchong hi~I have updated the raw dataset under this link: https://drive.google.com/drive/folders/1rIrgbF-0TY2_LI0a8WZZ0HoiYW6_3_PK?usp=sharing I process the data by"process_nerfstudio_to_sdfstudio.py",the command follow this: python scripts/datasets/process_nerfstudio_to_sdfstudio.py --data /data/yth/compare_work/studio_yth/sdfstudio-master/data_ours/gold_ring001/gold_ring001_raw --output-dir /data/yth/compare_work/studio_yth/sdfstudio-master/data_ours/gold_ring001/gold_ring001_process --scene-type object --data-type colmap

Have you also tried BakedSDF? I found it gives me better results for object-centric data. I would also try the parameter --pipeline.model.sdf-field.inside-outside False

hi,do you konw how to get a object-centric data? I have tried the BakedSDF with --pipeline.model.sdf-field.inside-outside False. It seems doesn't work.

[slpmee]

hi, I had troubled with same problem. try this. I added background-model mlp and background-color parameters.

ns-train neus-facto --pipeline.model.sdf-field.inside-outside False --vis viewer --experiment-name gold_ring_neuralangelo1 --data /data/yth/compare_work/sdfstudio-master/data_ours/gold_ring --viewer.websocket-port 7001 --trainer.steps-per-save 2000 --pipeline.model.sdf-field.bias 0.1 --pipeline.model.background-color white --pipeline.model.background-model mlp --vis viewer