[Models](/models)/[Captioning](/models?category=captioning)/facebook/Perception-LM-3B

HFScene CaptioningFAIR Noncommercial Research License

# Perception-LM-3B

by facebook

Meta Perception Language Model checkpoint for detailed image and video understanding

1.3Kdl/month

3B classparams

[HuggingFace](https://huggingface.co/facebook/Perception-LM-3B)[Run it on your own data, free](https://studio.mixpeek.com/namespaces/create?mode=managed&utm_source=homepage&utm_medium=model_page&utm_campaign=mvs_onramp)

Identifiers

Model ID

`facebook/Perception-LM-3B`

## Deploy Perception-LM-3B

Single-tenant

Mixpeek has no managed extractor for this model. On a single-tenant deployment you upload the weights and a custom plugin serves them next to the rest of your pipeline.

[Read the customer-hosted guide](https://mixpeek.com/docs/customer-hosted/overview) [Talk to us about single-tenant](/demo)

## Overview

Perception-LM-3B is part of Meta's PerceptionLM release for open, reproducible visual understanding research. The linked paper describes a transparent Perception Language Model stack for detailed image and video understanding, including human-labeled and synthetic data and a PLM-VideoBench evaluation for temporal perception.

On Mixpeek, Perception-LM-3B is useful when teams want a research-friendly VLM for building searchable descriptions of images and video clips. Its license is research-only, so it should be treated as an evaluation and prototyping model rather than a default commercial production choice.

## Architecture

Autoregressive vision-language model from the PerceptionLM family. The model combines a Perception Encoder visual backbone with a language decoder and is released in 1B, 3B, and 8B scales for detailed visual understanding experiments.

## Mixpeek SDK Integration

IngestRetrieve

// No extractor parameter takes a Hugging Face model id (checked against
// GET /v1/discovery/extractors, which returns 13), so Perception-LM-3B runs
// on your side and the output is upserted through POST
// /v1/namespaces/{namespace\_id}/documents/upsert. On Enterprise the other
// path is to upload the weights instead: POST /v1/namespaces/{id}/models
// accepts the huggingface format and a custom plugin loads them.
const res = await fetch(
  "https://api.mixpeek.com/v1/namespaces/ns\_your\_namespace/documents/upsert",
  {
    method: "POST",
    headers: {
      Authorization: "Bearer API\_KEY",
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      collection\_id: "col\_your\_collection",
      documents: \[
        {
          document\_id: "asset-00412",
          // The model produces text, so it lands in payload. Give the
          // collection a text vector index and embed that text to make it
          // searchable rather than only filterable.
          payload: { extracted\_text: modelOutput, source\_key: "archive/2026/asset-00412" },
          vectors: { "multimodal-embedding": embeddingOfModelOutput },
        },
      \],
    }),
  },
);

// Managed alternative, if this exact model is not the requirement:
// universal\_extractor@v1 runs google/gemini-embedding-2
// (3072-d) over a bucket, with no inference of your own.

## Capabilities

*   Detailed image and video understanding
*   Visual question answering over frames and clips
*   Temporal video perception research via PLM-VideoBench
*   Transparent data and training recipe for reproducible VLM evaluation
*   Useful baseline for comparing closed and open visual reasoning models

## Use Cases on Mixpeek

Prototype video understanding pipelines with an open research checkpoint

Compare caption quality across VLMs before selecting a production model

Index image and video datasets for agent evaluation

Build evidence traces for visual QA benchmarks

## Benchmarks

Dataset

Metric

Score

Source

PLM-VideoBench

Coverage

Introduced for temporal video understanding

PerceptionLM paper

Visual understanding tasks

Scope

Image and video understanding

HuggingFace paper page

## Performance

Input SizeImages or sampled video frames

GPU LatencyInput dependent

GPU ThroughputBatch dependent

GPU MemoryModel dependent

Research license requires access approval and noncommercial use

## Common Pipeline Companions

[

facebook/pe-av-large

Perception Encoder embeddings for cross-modal retrieval

](/model/facebook/pe-av-large)[

BAAI/bge-m3

Embed generated captions for text search

](/model/BAAI/bge-m3)[

DAMO-NLP-SG/VideoLLaMA3-7B

Compare video captioning quality across open VLMs

](/model/DAMO-NLP-SG/VideoLLaMA3-7B)

## Explore on Mixpeek

[

More Captioning models

Compare alternatives in this category

](/models?category=captioning)[

Best Document AI Platforms

Hand-picked tools & platforms compared

](/curated-lists/best-document-ai-platforms)[

OCR & Document AI Internals

Deep-dive technical guide

](/guides/ocr-document-ai-internals)[

Feature Extractors

See how Mixpeek runs models as extractors

](/extractors)[

Mixpeek Vector Store

Store & search embeddings at scale

](/mvs)[

Pricing

Usage-based pricing for pipelines

](/pricing)[

All Curated Lists

Compare models, APIs & infrastructure

](/curated-lists)

### Specification

FrameworkHF

Organizationfacebook

FeatureScene Captioning

Outputtext

Modalitiesvideo, image

RetrieverSemantic Search

Parameters3B class

LicenseFAIR Noncommercial Research License

Downloads/mo1.3K

[

### Research Paper

PerceptionLM: Open-Access Data and Models for Detailed Visual Understanding

arxiv.org](https://huggingface.co/papers/2504.13180)

### Build a pipeline with Perception-LM-3B

Add this model to a processing pipeline alongside other extractors. Combine with retrieval stages for end-to-end search.

[Run it on your own data, free](https://studio.mixpeek.com/namespaces/create?mode=managed&utm_source=homepage&utm_medium=model_page&utm_campaign=mvs_onramp)

### Alternative Models

[

Salesforce/blip2-opt-2.7b

Scene Captioning

](/model/Salesforce/blip2-opt-2.7b)[

microsoft/Florence-2-large

Scene Captioning

](/model/microsoft/Florence-2-large)[

google/paligemma2-3b-mix-448

Scene Captioning

](/model/google/paligemma2-3b-mix-448)[

Qwen/Qwen3-VL-8B-Instruct

Scene Captioning

](/model/Qwen/Qwen3-VL-8B-Instruct)