Perception Encoder
Meta's universal vision encoder for images, video, and audio.
About
Perception Encoder (PE) is Meta FAIR's family of vision encoders, released April 17, 2025 as part of the Perception Models project. It comes in core, language-aligned, and spatial variants covering images, video, and audio under one contrastive pretraining recipe. PE-lang can be plugged into vision-language models as the visual tower.
Meta claims PE core outperforms SigLIP 2 on image benchmarks and InternVideo2 on video, while PE spatial beats DINOv2 on dense prediction tasks. The weights are Apache-2.0 and the project is actively maintained, with integrations into timm and Hugging Face, and newer siblings such as PE-AV adding audio. It is purely a self-hosted research artifact with no official API.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Meta claims PE core outperforms SigLIP2 on image and InternVideo2 on video benchmarks; PE lang outperforms QwenVL2.5 and InternVL3 as a VLM visual tower; PE spatial outperforms DINOv2 on dense prediction (all provider claims, perception_models README)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified, metrics-v1). Evidence: Meta FAIR release 2025-04-17, facebookresearch/perception_models repo, HF collection, timm integration. Strengths: Apache-2.0 weights, active maintenance (PE-AV released Dec 2025), broad ecosystem adoption. Docked for: (1) one community analysis flagged an inconsistency between the repo's LICENSE.PE (Apache-2.0) and setup.py license metadata, worth checking before commercial embedding; (2) no official API. License: Apache-2.0 (PE weights; PLM sibling is non-commercial under separate terms).
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 9.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · Meta FAIR, released 2025-04-17; facebookresearch/perception_models
- · Integrated into timm and Hugging Face; PE-AV released Dec 2025
- · Provider claims beat SigLIP2/InternVideo2/DINOv2 on respective tasks