isarandi/nlf

[NeurIPS 2024] Neural Localizer Fields for Continuous 3D Human Pose and Shape Estimation

What it solves

Neural Localizer Fields (NLF) provides a method for estimating 3D human pose and shape from images or videos in a continuous manner. It addresses the challenge of accurately tracking human body movement and physical structure in 3D space.

How it works

The project uses a neural field-based approach to represent human body geometry and pose. By leveraging a continuous representation, the model can infer 3D human body parameters from visual input, with training code provided for both PyTorch and TensorFlow frameworks.

Who it’s for

This project is intended for researchers and developers working in computer vision, specifically those focused on human motion capture, 3D reconstruction, and embodied intelligence.

Highlights

  • Published at NeurIPS 2024.
  • Supports both PyTorch and TensorFlow implementations.
  • Includes training code for custom model development.
  • Designed for continuous 3D pose and shape estimation.

Related

  • Project
    PacesaLab/BindCraft2

    BindCraft2 is a Linux‑based command‑line pipeline that uses AlphaFold 2 and ProteinMPNN to design de novo protein binders, peptides, and antibody fragments for a user‑provided target. Users configure a JSON campaign (target, binder length, modality, optional properties) and the tool iterates through structure prediction, sequence redesign, and filter scoring, outputting ranked candidate sequences and predicted complex structures.

  • Project
    xibushu/UniForge3D

    UniForge3D is an AI system that turns a few unposed photos of an object into a complete 3D model. It jointly reconstructs observed geometry and generates missing parts in a shared latent space, using a view‑selection step to preserve details. The repo provides a Conda‑based install, a simple `infer.py` script, MIT‑licensed code, and pretrained weights (subject to their original licenses).

  • Project
    jasongzy/Make-It-Animatable

    An efficient framework that automates the creation of animation-ready 3D characters by predicting joint positions and skinning weights from 3D meshes.

  • Project
    Dai-Wenxun/MotionLCM

    MotionLCM is a research codebase for real‑time, text‑conditioned human motion generation. It uses a latent consistency model for one‑step inference and a Motion ControlNet to incorporate explicit joint trajectories, supporting motion reconstruction, text‑to‑motion, and controllable motion synthesis. The README provides conda setup, data preparation (HumanML3D), pretrained model download, demo commands, rendering with SMPL/Blender, and full training instructions via YAML configs.