Hacktoberfest 2026: the issues maintainers tagged for October, open and beginner-friendly. Browse Hacktoberfest issues

slat_decoder_mesh failed.

Open
  • Is something wrong?
#166 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
25/100
Issue type
Bug
Clarity
Needs clarification
Activity status
Stale
Tech stack
python, pytorch
Domain
machine-learning

Research direction

Start at decode_slat in the reported model path and reproduce the call to self.models"slat_decoder_mesh" using the NVIDIA GeForce RTX 4090 setup. Compare the mesh decoder forward pass with the working gaussian path and trace where NaN values first appear in SLatMeshDecoderTdfyWrapper. Done means identifying the cause of the mesh-only NaNs and confirming that output["glb"] is no longer empty.

Written by the indexing model from the issue text.

Description

Question description

I'm able to access and visualize outputs["gs"] correctly via the supersplat editor, but output["glb"] returns empty. Tracing this back to the decode_slat function:

device: NVIDIA GeForce RTX 4090 (48GB)

  def decode_slat(
        self,
        slat: sp.SparseTensor,
        formats: List[str] = ["mesh", "gaussian"],
    ) -> dict:
        """
        Decode the structured latent.

        Args:
            slat (sp.SparseTensor): The structured latent.
            formats (List[str]): The formats to decode the structured latent to.

        Returns:
            dict: The decoded structured latent.
        """
        logger.info("Decoding sparse latent...")
        ret = {}
        with torch.no_grad():
            if "mesh" in formats:
                ret["mesh"] = self.models["slat_decoder_mesh"](slat)
            if "gaussian" in formats:
                ret["gaussian"] = self.models["slat_decoder_gs"](slat)
            if "gaussian_4" in formats:
                ret["gaussian_4"] = self.models["slat_decoder_gs_4"](slat)
        # if "radiance_field" in formats:
        #     ret["radiance_field"] = self.models["slat_decoder_rf"](slat)
        return ret

the issue occurs at ret["mesh"] = self.models"slat_decoder_mesh". Going further, I found that NaN values appear during the forward pass of
'sam3d_objects.model.backbone.tdfy_dit.models.structured_latent_vae.decoder_mesh.SLatMeshDecoderTdfyWrapper'. I have checked the model weight loading, and the weights are loaded correctly." Has anyone encountered similar issues? Any suggestions on what might cause NaN only in the mesh decoder?

Dominant language
Python
Stars
7.5k
Forks
891
No merged PRs in 30d

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from facebookresearch/sam-3d-objects

All issues in facebookresearch/sam-3d-objects

Similar issues

More Python issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.