gitwtfhub

wtf is sapiens?

facebookresearch/sapiens — explained in plain English

Analysis updated 2026-07-03 · repo last pushed 2026-05-26

★ 5,393PythonAudience · researcherComplexity · 4/5MaintainedSetup · hard

TL;DR

A set of AI models trained on 300 million human images that can detect body pose, segment body parts, and estimate depth from high-resolution real-world photos and videos.

Mindmap

mindmap
  root((sapiens))
    What it does
      Detect body pose
      Segment body parts
      Estimate depth
      Surface normal estimation
    Training Data
      300M human images
      Real-world diversity
      1024x1024 resolution
    Use Cases
      Fitness apps
      Fashion and tailoring
      Animation and gaming
      Accessibility tools
    Tech Stack
      Python
      Pretrained models
      Fine-tuning support
    Recognition
      CVPR 2024 best-paper candidate

Code map

Detail Auto

An interactive map of this repo's files and how they connect — its source is parsed live in your browser. Click Visualize to build it.

filefunction / class

Why would anyone build with this?

REASON 1

Track a person's exercise form in real time by running Sapiens pose detection on video frames from a fitness app.

REASON 2

Estimate garment fit and body proportions in fashion photos using Sapiens body-part segmentation.

REASON 3

Capture human motion for a game character by extracting joint positions from ordinary webcam footage with Sapiens pose detection.

REASON 4

Fine-tune one of the Sapiens models on your own dataset of specialized body poses for a medical or sports analytics application.

What's in the stack?

PythonPyTorch

How it stacks up

facebookresearch/sapienskarpathy/build-nanogptkarpathy/neuraltalk
Stars5,3935,3055,495
LanguagePythonPythonPython
Last pushed2026-05-262024-08-132020-12-22
MaintenanceMaintainedStaleDormant
Setup difficultyhardmoderatehard
Complexity4/53/54/5
Audienceresearcherdeveloperresearcher

Figures from each repo's GitHub metadata at analysis time.

How do you spin it up?

Difficulty · hard Time to first run · 1h+

Requires GPU for practical inference, full install needed for fine-tuning, lightweight install available for inference-only use.

No license information is mentioned in the explanation.

Yoink these prompts

Prompt 1
Using the Sapiens pose detection model, write Python code that loads a JPEG image of a person and prints the (x, y) coordinates of all detected body keypoints.
Prompt 2
Help me set up Sapiens body part segmentation on a folder of fitness videos: process each frame, overlay the segmentation mask, and save the annotated video.
Prompt 3
I want to fine-tune the Sapiens depth estimation model on my own dataset of indoor climbing photos. Walk me through the training setup and data format required.
Prompt 4
Show me how to run Sapiens surface normal estimation on a 1024x1024 portrait photo and visualize the output as a color-coded normal map.
Prompt 5
I'm building a tailoring app that needs to measure shoulder width and torso length from a front-facing photo. How do I use Sapiens keypoints to calculate these measurements in pixels and convert them to real-world centimeters?

Frequently asked questions

wtf is sapiens?

A set of AI models trained on 300 million human images that can detect body pose, segment body parts, and estimate depth from high-resolution real-world photos and videos.

What language is sapiens written in?

Mainly Python. The stack also includes Python, PyTorch.

Is sapiens actively maintained?

Maintained — commit in last 6 months (last push 2026-05-26).

What license does sapiens use?

No license information is mentioned in the explanation.

How hard is sapiens to set up?

Setup difficulty is rated hard, with roughly 1h+ to a first successful run.

Who is sapiens for?

Mainly researcher.

View the repo → Decode another repo

This repo across BitVibe Labs

Don't trust strangers blindly. Verify against the repo.

Other repos in this lane