open source · vision · accessibility
        ◦ ─── +
       /       \
      ·         ◦ ── ·
     / \         \
    +   ◦ ─── ·   +
         \       /
          · ── ◦

See the world.
Every pixel.
For everyone.

Sarvadrushti is an open-source vision and accessibility toolkit — built to make digital environments navigable for people of all visual abilities.

View on GitHub Learn more
2024
founded
OSS
license
India
origin
∞
reach
about

Built as a startup.
Released for everyone.

+---------------------------+
|  SARVADRUSHTI             |
|  "all-seeing"             |
|                           |
|  ·  vision assistance     |
|  ·  screen accessibility  |
|  ·  navigation tools      |
|  ·  real-time processing  |
|                           |
|  started: 2024            |
|  status:  open source     |
|  license: MIT             |
+---------------------------+

Sarvadrushti ("all-seeing" in Sanskrit) started as a pre-incubated startup at VNRVJIET, Hyderabad, in February 2024 — with a single goal: close the accessibility gap in digital tools for the visually impaired.

After a year of building, we opened the source. Everything we built — models, pipelines, interfaces — is now yours to use, fork, and improve.

Backed by the belief that accessibility is not a feature, it's a foundation. 

$ git clone https://github.com/Project-Sarvadrushti/Sarvadrushti
Cloning into 'core'...
remote: Enumerating objects: 1,204
$ pip install sarvadrushti
Successfully installed sarvadrushti
$ sarva --help
Usage: sarva [command] [options]
  describe  ·  enhance  ·  navigate  ·  audit
features

What it does.

A modular toolkit — use individual components or the full suite. Everything ships with a Python API and a CLI interface. 

[ >> ]
Real-time Scene Description
Converts live camera feed or static images into structured natural language descriptions, tuned for low-latency assistive use. 
[ ## ]
Contrast & Layout Auditing
Scans UI screens or web pages for WCAG compliance issues — contrast ratios, font sizing, tappable target sizes — and outputs actionable fixes. 
[ ~~ ]
Navigation Assistance
Spatial audio and haptic-feedback guidance overlays for navigating physical and digital environments with minimal vision. 
[ :: ]
Text & Document Enhancement
Adaptive readability pipelines — reformats dense documents into screen-reader-optimized formats with semantic structure intact. 
[ ** ]
Modular Python API
Every component is an importable Python module. Compose your own accessibility stack with minimal boilerplate. 
[ ++ ]
Multilingual Support
Built with Indian language support from day one — Hindi, Telugu, Tamil, and more. Because accessibility is local.
downstream evaluation

Vision that understands.

Sarvadrushti's dehazing pipeline restores degraded frames end-to-end — improving not just visual quality, but downstream task performance across object detection and scene captioning.

[ clear ] YOLOv11 detects bicycle (0.84) and bottle (0.69) with high confidence. Captioning accurately identifies the storage environment, easel, and painting.
[ hazy ] Detection fails under dense haze. Captioning degrades — missing key objects and misattributing lighting and depth cues.
[ dehazed ] Post-restoration, detection recovers — bicycle at 0.97 confidence. Captioning restores full scene fidelity with accurate object placement.