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

Pointnetvlad

Pointnetvlad is an open-source deep-learning model and research codebase that retrieves point-cloud submaps for large-scale place recognition. It includes benchmark datasets and scripts to prepare training and test files, train baseline and refined networks, and evaluate a model, as well as downloadable pretrained models.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Pointnetvlad — 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 it when a research workflow needs point-cloud-based place recognition and the benchmark preparation, training or evaluation steps documented in the repository.

Who it is for

  • Robotics and vehicle-mapping researchers working on point-cloud place recognition.
  • Researchers who need benchmark data and scripts for training or evaluating point-cloud retrieval models.

Documented uses

  • Retrieve matching point-cloud submaps to recognize places.
  • Prepare positive and negative point-cloud examples from benchmark data.
  • Train baseline or refined networks and evaluate them on benchmark test data.
  • Use the provided pretrained baseline or refined models in a point-cloud retrieval experiment.
Before you choose

Requirements & limits

The repository is public and released under the MIT License. The documented environment uses Python 3, TensorFlow 1.4.0 and CUDA 8.0.

  • Benchmark inputs are binary submaps paired with CSV metadata; the CSV files define positive and negative examples.
  • Benchmark submaps are road-removed and downsampled to 4,096 points.
  • Oxford and NUS dataset layouts and train/test splits differ; the README specifies corresponding folders and CSV files.
  • The README lists CUDA and TensorFlow among prerequisites and reports testing with CUDA 8.0 and TensorFlow 1.4.0.
  • The repository provides separate baseline and refined network preparation workflows and pretrained models.

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 benchmark files for training or evaluation

Generate the pickle files used by Pointnetvlad's training or evaluation workflows.

Before you start

  • Python 3 environment with the listed Python, CUDA, TensorFlow, SciPy, Pandas and scikit-learn prerequisites.
  • The benchmark dataset archive extracted alongside the project code so both benchmark_datasets/ and pointnetvlad/ are present.

Steps

  1. Open the generating_queries/ directory.

  2. Run generate_training_tuples_baseline.py for baseline-network training tuples, or generate_training_tuples_refine.py for refined-network training tuples.

  3. For network evaluation data, run generate_test_sets.py.

Expected outcome

Pickle files containing the positive and negative point clouds for anchors, prepared for the selected training or evaluation workflow.

Guide limitations

  • The README says tuple generation is needed only once and may take a few minutes.
  • The repository documents Python 3, TensorFlow 1.4.0 and CUDA 8.0 as the environment used for the code; this is not a compatibility guarantee.
Documentation 1 ↗
Evidence and transparency

Sources & review

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