Sangluisme / PSgradientSDF

Hi , I have a sequence of RGBD images . But the result of reconstruction seems only contain the first image. Could you please give some suggestions?

My config:

{
    "input": "/mnt/C07F201295383FC7/Data/berxekl/box_full_png_PSGSDF/", 
    "output": "/mnt/C07F201295383FC7/Data/berxekl/box_full_png_PSGSDF/output3/",
    "datatype": "synth",
    "first": 0,
    "last": 200,
    "voxel size": 0.001,
    "truncation factor": 5,
    "zmin": 0.0,
    "zmax": 3.5,
    "sharpness threshold": 0.03,
    "model type": "LED",
    "loss function": "cauchy",
    "reg albedo": 0.0,
    "reg norm": 0.1,
    "reg laplacian": 5.0,
    "max iter": 100,
    "damping": 3.0,
    "converge threshold": 5e-3,
    "lambda": 0.2,
    "upsample":true,
    "--light": true,
    "--albedo": true,
    "--distance": true,
    "--pose": false
}

some of my rgb data:

I saw you set "--pose false", which means you need to specify the poses of the sequence. Do you have the poses of the camera already? If not, the set "--pose true". You can try with "--upsample false" at very beginning. And if you use the data type "synth", that means your data format should be as described in README.md.

And according to this:

PSgradientSDF/cpp/include/img_loader/SynthLoader.h

Line 24 in ee5bb02

ImageLoader(1./1000, true),

The depth factor is 1000, which means the depth value divide by 1000, is the true distance. Please make sure your data format satisfies this.

Thanks for your reply.
I found the reason. This is because the initialized griddim is (128, 128, 128). since I set voxel size=0.001, the volume is too small. After change it to (640, 640, 640), it worked.
BTW, I think parameter voxel_size is in m , not cm. Because my box is around 32cm length and 25cm width. If the unit is cm , the volume is around 0.64³ cm³, which cannot cover my box. But if it is m, I think this volume 64³ cm³ can cover the box.

Yes, the unit for voxel size is meter. Sorry if we write it is cm. It should be correct in the paper. I'll check if the readme.md is also correct.

I corrected the readme.md and add two trouble shooting for this problem. Sorry for the mistakes.

Thanks!

Question about output mesh only contains one image