SShortSingh.
Back to feed

Procedural Gaussian splats in Houdini: six fits show gains from 12 to 24 training views

0
·6 views

A developer rendered a small procedural valley scene in Houdini and trained Gaussian splats on the synthetic images produced. Using 4,096 splats and 600 training steps, raising the training views from 12 to 24 improved held-out quality for both seeds tested. Moving from 24 to 36 views added little full-frame PSNR and slightly lowered foreground PSNR. The findings apply only to one 192 by 192 scene and a fixed-capacity trainer, with fit times and memory figures reported for this tiny setup. The 36-view seed-381 result was chosen for delivery, and its compressed SPZ file was 76.7 percent smaller than the binary PLY while still loading in a browser viewer.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Granting One User Write Access Without chmod 777: Using POSIX ACLs on Linux

A common shortcut for giving a deploy user write access to a single directory is chmod 777, which exposes the files to every account on the host. Adding the user to the directory's group is another option, but it grants that group's access everywhere it applies and may not take effect in an already-running session. POSIX ACLs offer a per-object alternative that grants permissions to a named user or group without changing ownership or opening access to everyone. The article explains that ACLs add named entries alongside the traditional owner, group, and other bits, with a mask entry acting as a ceiling on named users and groups. It walks through setting up a setgid shared directory and granting a deploy user access with setfacl, noting that the capital X flag grants execute only to directories.

0
ProgrammingDEV Community ·

LiteLLM 1.107.0 on PyPI and Docker Carried Malware After Leaked Admin Key Was Misused

An administrative key for an internal LiteLLM Agent Platform deployment was accidentally committed to a public repository on May 27, 2026, and deleting the file three hours later did not revoke it. In October, an attacker used that key to pull configuration data containing a GitHub CLI token belonging to a maintainer of LiteLLM's GitHub organization. With that token, the attacker gained owner-level access, ran workflows that exfiltrated more maintainers' tokens, and pushed a change to the release pipeline. On October 21, malicious litellm 1.107.0 wheels were published to PyPI and matching Docker tags were pushed, and they remained available for about two hours before removal. LiteLLM said the malicious files were never part of its official v1.107.0 source tag, and Wiz estimated roughly 49,000 downloads during that window.

0
ProgrammingDEV Community ·

Developer Reflects on Daily Painkiller Reliance While Coding and Its Health Risks

A developer writing on DEV Community says that, as they have gotten older, they have come to rely on Loxonin, a painkiller, almost every day while working on software. They note they have heard that taking too much of the drug is harmful. The post offers a personal account rather than a formal study or medical guidance. No specific dosage figures or medical conclusions are provided in the item.

0
ProgrammingDEV Community ·

pam_exec environment data is not a safe channel for secrets

The article warns that pam_exec, a PAM module that runs external commands during authentication stages, exposes PAM items as environment variables that can be influenced by clients or service configuration. Scripts that parse these values as shell fragments, log them insecurely, or use them in command lines may introduce vulnerabilities. The expose_authtok option is especially risky because it passes the authentication token to the called command, which could leak credentials if the command is writable or insecure. The author recommends administrator-owned helpers, fixed executables, protected logs, and dedicated file descriptors or protected temporary files for secrets. Thorough testing of failure paths is also advised, since the PAM control flags determine whether errors accidentally allow authentication.