OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Mvgformer

MVGFormer is an open-source research implementation of a transformer model that estimates 3D human poses from multiple camera views by combining image-feature attention, multi-view geometry and triangulation.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Mvgformer — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use MVGFormer when you need a research codebase for training or evaluating transformer-based 3D human pose estimation from multiple camera views, rather than a hosted inference service.

Who it is for

  • Computer-vision researchers studying multi-view 3D human pose estimation.
  • Developers prototyping pose pipelines that use multiple calibrated camera views.

Documented uses

  • Train a multi-view 3D pose model using the CMU Panoptic dataset.
  • Evaluate pose estimation under changes in camera count, camera arrangement or dataset.
  • Run documented inference on Shelf and Campus datasets.
  • Adjust model configuration parameters for ablation experiments.
Before you choose

Requirements & limits

The repository is public and specifies the Apache-2.0 license.

  • The documented environment uses Python 3.10, Conda, MMCV and CUDA-compiled attention modules.
  • CMU Panoptic preparation requires video, calibration and 3D body keypoint data, plus a separate PoseResNet-50 checkpoint.
  • The README's CMU Panoptic training command uses eight GPUs.
  • The repository documents generalization evaluation and inference support for Shelf and Campus datasets.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Prepare the MVGFormer environment and Panoptic data

Set up the documented software environment and required CMU Panoptic inputs for the repository.

Before you start

  • A Linux environment with Python 3.10 and Conda.
  • CUDA configured for MMCV and the attention-module build.
  • CMU Panoptic data, calibration files and 3D body keypoints.
  • The PoseResNet-50 pretrained checkpoint.

Steps

  1. Clone the MVGFormer repository and create and activate a Conda environment with Python 3.10.

  2. Install MMCV using OpenMIM with attention to the CUDA version, then install the repository requirements.

  3. Set CUDA_HOME and build the attention modules from ./lib/models/ops.

  4. Place the CMU Panoptic data under ${POSE_ROOT}/data/panoptic/ and the pretrained checkpoint at ${POSE_ROOT}/models/pose_resnet50_panoptic.pth.tar.

Expected outcome

The documented software, dataset and pretrained-backbone inputs are prepared for the repository's subsequent training or evaluation stages.

Guide limitations

  • This recipe covers setup only; it does not verify a successful build or model run.
  • The README directs users to external dataset and pretrained-model resources.
  • CUDA and MMCV installation must be matched to the environment.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
16 October 2026
Original website title
XunshanMan/MVGFormer · Official homepage title · retained in its source language
How we review sourcesSuggest a correction