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

Frustum Pointnets

Frustum PointNets is an open-source Python research repository implementing 3D object detection from RGB-D data. It combines 2D image detections with point-cloud processing to perform 3D instance segmentation and estimate amodal 3D bounding boxes.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Frustum Pointnets — 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 you want a research implementation that links RGB image detections to point-cloud-based 3D segmentation and box estimation, and can work with a local machine-learning setup.

Who it is for

  • Researchers studying 3D object detection and point-cloud learning.
  • Developers building or adapting robotics perception research prototypes.
  • Teams evaluating 3D detection methods on KITTI or working with the repository’s supported data workflow.

Documented uses

  • Train and test Frustum PointNets models.
  • Prepare KITTI image and point-cloud data as frustum-based training examples.
  • Evaluate detection results using precomputed 2D detector outputs.
  • Adapt the codebase for another 2D detector or a custom dataset.
Before you choose

Requirements & limits

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

  • The pipeline depends on both RGB image detections and corresponding 3D point clouds.
  • The documented environment centers on TensorFlow 1.2 or 1.4 and Ubuntu 14.04 or 16.04; Python 2.7 is listed, while Python 3 is described as potentially workable.
  • The README recommends GPU access and requires at least 8 GB of available CPU memory.
  • Using the v2 model requires compiling custom TensorFlow operators; average-precision evaluation requires compiling separate evaluation code.
  • The KITTI preparation command generates about 4.7 GB of data.

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 KITTI data for Frustum PointNets

Convert KITTI images, Velodyne point clouds, calibration and labels into the repository’s training data format.

Before you start

  • KITTI 3D object detection data: left color images, Velodyne point clouds, camera calibration matrices and training labels.
  • Arrange the dataset as described in the repository’s dataset/README.md.
  • Expect approximately 4.7 GB of generated pickle files.

Steps

  1. Download the KITTI 3D object detection dataset and organize it according to dataset/README.md.

  2. Run the command below to inspect whether the data is stored properly.

    Documented command
    python kitti/kitti_object.py

  3. Run the command below to extract frustum point clouds and labels.

    Documented command
    sh scripts/command_prep_data.sh

Expected outcome

The preparation process generates three data files under the kitti folder for the training and validation workflow.

Guide limitations

  • This procedure prepares KITTI data; it does not cover model training or inference.
  • The generated pickle files require around 4.7 GB of storage.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
16 October 2026
Original website title
charlesq34/frustum-pointnets · Official homepage title · retained in its source language
How we review sourcesSuggest a correction