SimonGiebenhain / MonoNPHM

49 stars 8 forks source link

Reconstruction from custom data failed #3

Open yuhongyun777 opened 1 week ago

yuhongyun777 commented 1 week ago

Thank you for your efforts! While running your code with the demo data, everything works perfectly; the mesh reconstruction aligns well with the expressions and shape of the demo data.

However, when I attempt to use my own data following your preprocessing steps, the reconstructed results seem very odd and inaccurate. I have reviewed the visualizations from the metrical-tracker and found no apparent issues. The segmentation and keypoint detection results also appear normal.

The only potential issue seems to be with the intrinsic and extrinsic parameters obtained from the metrical-tracker. Specifically, the focal length in the intrinsic parameters is significantly larger compared to the demo data, and the translation values in the extrinsic parameters are much smaller.

Could you please advise on what might be causing this problem?

Additionally, I utilized your preprocessing code to preprocess the demo data (00059) again. However, the reconstruction results are incorrect, similar to the issues I faced with my custom data. Therefore, I suspect the problem might lie within the preprocessing code or my environment setup. Could you provide some guidance on this matter?

reconstruction from demo_data 00059:

image

reconstruction from my own data:

image

metrical tracker results from my own data: image

intrinsic and extrinsic diff:

3A7B156B-F612-4DB7-B6F6-4E98AC033094
SimonGiebenhain commented 6 days ago

Hi, Is it possible that you ran bash run.sh SEQ_NAME --intrinsics_provided?

The —intrinsics_provided flag will ignore the estimated intrinsics from MICA and use the intrinsics from our test camera instead.

If that is the case, can you try again with —no-intrinsics_provided?