Picture of the author

Build computer vision models for all use cases quickly

Synativ is a toolbox which leverages generative AI and foundation models to help computer vision teams build models for all use cases quickly without the need to collect & label vast amounts of images.

synativ-demo.py
requirements.txt
from synativ.api import Synativ
from synativ import Model
synativ_api: Synativ = Synativ(api_key="{YOUR_API_KEY}")
model: Model = synativ_api.fine_tune(
base_model="sam_semantic_big",
dataset_id="<dataset_id>",
metadata={}
)

Getting Started

Quickstart

We believe that the future of computer vision centers around hierarchical fine-tuning starting from domain-agnostic foundation models trained on a large amount of diverse data, to domain-specific foundation models trained on a large amount of data of a particular domain, to application-specific models trained on a small amount of data for a particular use case.

Using Synativ you can start working with visual foundation models in a few lines of code instead of wrangling for days with GitHub repositories belonging to papers. In the background, we are constantly improving the performance of our fine-tuning algorithms and domain-specific starting points.

You can get started with Synativ in 3 simple steps:

  1. Install Synativ's SDK
  2. Fine-tune a foundation model
  3. Start hosting your fine-tuned model

0. Prerequisite: Request Your API Key

To receive your personal API key, enter your email address below and click the button:

1. Install Synativ's SDK

Synativ SDK is not yet publicly available but you can install it using pip:

pip install https://synativ-sdk-early-version.s3.amazonaws.com/prod.zip

2a. Fine-Tune a Domain-Agnostic Foundation Model

Pick this option if you have a large amount (100k+) of images available to create your proprietary, domain-specific foundation model by fine-tuning a domain-agnostic foundation model. Your images can be labelled or unlabelled.

You can later fine-tune your proprietary foundation model for custom applications with a smaller amount of data, thereby drastically decreasing the iteration time and cost to deploy a new use case.

Segment Anything (SAM)

A generic segmentation model with zero- and few-shot generalization to unseen objects and images from within its training domain.

EfficientSAM

A light-weight SAM model that exhibits decent performance with largely reduced complexity for applications with constrained resources.

2b. Fine-Tune a Domain-Specific Foundation Model

Pick this option if you do not have a large amount of images available. We have prepared a few domain-specific foundation models as a starting point for you. As we have trained them on a large amount of images of the respective domains, you only need a small amount of data to fine-tune them for your application.

Let us know if your domain is missing as we are constantly training new models!

Agriculture

Synativ's foundation model for developing agriculture applications.

Geospatial

Synativ's foundation model for developing geospatial applications.

Manufacturing

Synativ's foundation model for developing manufacutring applications.

Pathology

Synativ's foundation model for developing pathology applications.

3. Start Hosting Your Fine-Tuned Model

After your model has been fine-tuned, you can start hosting it for real-time inference.