OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Image & Video AI

Lighthouse

Lighthouse is an Apache-2.0 Python library for retrieving text-query-relevant moments in video and audio, detecting video highlights, and reproducing media-retrieval workflows through inference APIs and demos.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Lighthouse — OpenViST project profile in Image & Video AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Lighthouse when building or reproducing a Python-based workflow that maps text queries to relevant video or audio segments, or that combines video moment retrieval with highlight detection.

Who it is for

  • Researchers reproducing video moment-retrieval, highlight-detection or audio-retrieval workflows.
  • Developers prototyping text-based search across video or audio recordings.

Documented uses

  • Find time ranges in a video that match a natural-language query.
  • Identify salient segments in a video for highlight selection.
  • Search audio recordings for moments matching a text description.
  • Reproduce example video or audio inference workflows using the provided demo scripts.
Before you choose

Requirements & limits

The repository is public and identifies the project as open source under the Apache-2.0 license. Its README provides installation and local demo instructions.

  • Video inference returns relevant time windows and saliency scores; audio retrieval uses a separate AMR predictor and audio encoding path.
  • The README describes seven models, four features and six datasets for its reproducible retrieval and highlight-detection workflows.
  • Video input is limited to 150 seconds according to the README.
  • CPU inference is supported, but the README warns that SlowFast-based feature combinations are very slow without GPUs and recommends the CLIP feature for CPU use.
  • Some feature configurations require additional pretrained weights, including SlowFast weights and, for CLIP+SlowFast+PANNs, PANNs weights.

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

Set up Lighthouse and run the documented video demo

Install the documented dependencies and run the repository’s video moment-retrieval and highlight-detection demo.

Before you start

  • Python environment; the repository reports testing with Python 3.9 and CUDA 11.8.
  • ffmpeg.
  • PyTorch, torchvision and torchaudio installed for the intended GPU environment; CPU inference is also supported.
  • Access to the Lighthouse GitHub repository.

Steps

  1. Install ffmpeg; the README gives apt install ffmpeg for Ubuntu.

  2. Install PyTorch, torchvision and torchaudio appropriate to the GPU environment. The README provides a CUDA 11.8 example using PyTorch 2.1.0, torchvision 0.16.0 and torchaudio 2.1.0.

  3. Install Lighthouse with the command below.

    Documented command
    pip install 'git+https://github.com/line/lighthouse.git'

  4. Run the command below to reproduce the video moment-retrieval and highlight-detection results; the script automatically downloads pretrained weights.

    Documented command
    python api_example/demo.py

Expected outcome

The documented video demo runs the MR-HD example and produces query-relevant time windows and saliency scores.

Guide limitations

  • The README states that video duration is limited to 150 seconds.
  • On CPU, the README recommends feature_name='clip'; CLIP+SlowFast and CLIP+SlowFast+PANNs are described as very slow without GPUs.
  • The listed test environment is Python 3.9 and CUDA 11.8; this is not a compatibility guarantee.
Documentation 1 ↗
Source-based practical guide

Run the documented audio moment-retrieval demo

Set up the dependencies and run Lighthouse’s audio moment-retrieval example.

Before you start

  • Python environment.
  • ffmpeg.
  • PyTorch, torchvision and torchaudio installed for the intended GPU environment; the inference API is also available for CPU mode.
  • Access to the Lighthouse GitHub repository.

Steps

  1. Install ffmpeg; the README gives apt install ffmpeg for Ubuntu.

  2. Install PyTorch, torchvision and torchaudio appropriate to the environment.

  3. Install Lighthouse with the command below.

    Documented command
    pip install 'git+https://github.com/line/lighthouse.git'

  4. Run the command below; the README says the demo automatically downloads pretrained weights.

    Documented command
    python api_example/amr_demo.py

Expected outcome

The documented AMR demo reproduces an audio moment-retrieval example, matching a text query to relevant moments in audio.

Guide limitations

  • The README recommends CPU use with the clip feature for its inference API; it does not specify an audio-duration limit in the cited setup instructions.
  • The example’s automatic weight download is described for the demo script; other model or feature choices can require separately downloaded weights.
Documentation 1 ↗
Evidence and transparency

Sources & review

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