Data Preparation

[sinAshish]

Hi @EnyaHermite, thanks for open-sourcing the code for this awesome work.
Could you please provide more information on how to process the datasets, such as scannet, scannet_texture, etc? I'm trying to run the pytorch branch of this repo.

Moreover, the gdrive link for the trained models is not working?

[EnyaHermite]

I am uploading the data to google drive, which is slow due to their large size. Should be ready next week sometime. Thank you very much for the interest.

[sinAshish]

Thank you, @EnyaHermite !

[sinAshish]

Hi @EnyaHermite, are there examples on how to prepare our own 'textured/rendered' data as per your pytorch datasets?

[UncleSan]

Hi! @EnyaHermite, could you please tell me how to transform the seg-labels of facset to vertexes in a mesh? When I used your Picasso-II to train my own data, the acc. is only 4.12% and miou 0.5. I just want to make sure if I were wrong in preparation step. Thank you!

[sinAshish]

Hi @EnyaHermite, could you please describe how to process our own datasets to get these:

baryPath = meshPath.replace('vh_clean_2.ply', 'rendered_baryCoeffs.txt')
    baryCoeff = np.loadtxt(baryPath, delimiter=',', dtype=np.float32)
    if visualize:
        baryCloud = o3d.geometry.PointCloud()
        baryCloud.points = o3d.utility.Vector3dVector(baryCoeff)
        o3d.visualization.draw_geometries([baryCloud],mesh_show_back_face=True)
    # =================================================================================

    # =============================location normalization==============================
    numinteriorPath = meshPath.replace('vh_clean_2.ply', 'rendered_numInterior.txt')
    numTexture = np.loadtxt(numinteriorPath, dtype=np.int32)
    ```

[assissanliu]

hello, Have you had any problems when you installing picasso?