Skip to content

datasaurus-ai/datasaurus

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

27 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

logo

Datasaurus

Do computer vision with 1000x less data

License Apache-2.0

Hosted App (coming soon) - Running Locally


Leverage a foundational text-vision model for your computer vision tasks. Instead of having to train your own models from scratch, rely on pre-trained models. You can achieve great performance with no data, and an even better one with a couple of datapoints. If you already have a lot of data, then exceed your previous models' performance by fine-tuning a foundational model (coming soon).

demo

Features

  • Fully open-source
    • Fine-tuned model weights can be downloaded.
  • Do computer vision tasks without any data.
  • Promptable system
    • If your requirements change, then you just need to adjust your prompt; no need to retrain an entire computer vision model.
  • And many more features coming soon...

Examples

Instead of training a model from scratch, you can just prompt your images.

  • Color Detection Pipeline
    • Prompt: Determine the main color of specific objects within an image.
  • Count and Action Recognition Pipeline
    • Prompt: Identify the number of people in a scene and their actions.
  • Fruit Ripeness Analysis Pipeline
    • Prompt: Analyze images of fruit to determine their level of ripeness.
  • Dog Breed Identification Pipeline
    • Prompt: Classify the breed of a dog from a given image.

Supported Base Models

  • LLaVA-v1.5-7B
  • LLaVA-v1.5-13B
  • GPT4-V (as soon as the model API is available)

Roadmap

  • v0 launched
  • Add examples sections.
  • Dataset importer + associated dashboard.
  • Parameterizable prompt.
  • Fine-tune models.
  • Add visual in-context learning.
  • Support for additional inference backend (gglm).
  • Hosting service deployment (right now, waitlist).
  • Stronger output guidance.

Running Locally

  1. Install NodeJS 20 (earlier versions will very likely work but aren't tested)
  2. Install Supabase with npm i supabase --save-dev
  3. Install Conda
  4. Clone this repository and open it: git clone https://github.com/datasaurus-ai/datasaurus && cd datasaurus
  5. Install the frontend dependencies: cd frontend && npm install && cd ..
  6. Install the backend dependencies: cd backend && virtualenv datasaurus-backend && source datasaurus-backend/bin/activate && pip install -r requirements.txt && cd ..
  7. Start Supabase: cd supabase && supabase start && cd ..
  8. Register for an account on replicate.com and obtain an API key. We utilize Replicate as our backend for performing inference. (Note: running models locally capability coming soon.)
  9. Create the backend .env file (cd backend && cp .env.example .env && cd ..) and complete it
  10. Create the frontend .env file (cd frontend && cp .env.example .env && cd ..) and complete it
  11. Start the backend: cd backend && source datasaurus-backend/bin/activate && uvicorn src.main:app --reload && cd ..
  12. Start the frontend: cd frontend && npm run dev && cd ...
  13. Navigate to http://localhost:3000

Interested?

If you are interested, please leave us a star and/or sign up for launch of the hosted version on datasaurus.app