Picarta is the first startup that tries to solve the problem of image geolocalization, which
is finding where a photo has been taken in the world. Our mission at Picarta is to verify
photo locations worldwide, ensuring authenticity, countering misinformation, and offering
reliable geolocation solutions. 🌍🔍
We began our journey in August 2023 with the launch of Picarta V1. Our initial release
quickly garnered the interest of AI researchers, businesses, government entities, and OSINT
communities around the globe.
After a year of dedicated research and development, we are excited to announce the launch of
Picarta V2, the most accurate and reliable image geolocalization model in the world. Our AI
model is based on vision transformers (ViTs), which have shown exceptional performance
across various applications in computer vision. To give you an idea of our model’s accuracy,
we compared it with state-of-the-art models like ISNs, GeoDecoder, GeoCLIP and PIGEOTTO.
Picarta V2 scores 30.4% accuracy at a 1km distance on the IM2GPS3k dataset. However, it’s
important to note that this comparison might not be entirely fair since the experimental
conditions differ significantly between Picarta V2 and other methods. We mention this to
provide a benchmark for our model’s accuracy.
We estimate with a probability of 70% that our model can predict the GPS location of any
outdoor image by looking at the Top 100 predictions returned by our AI model for the input
image. In other words, if the user investigates the Top 100 results returned by our model,
there is a 70% chance of finding the location of the input image, which is an unprecedented
achievement in this field!
Of course, the user can also further boost performance by providing the AI with prior
knowledge about the location. This can be done by specifying the location by drawing a
circle on the map of the area where the user believes the image was taken.
How to use Picarta?
The Picarta model can be used in three different ways: through our website, via API, or using an offline model.
Website: On our website, users can upload an image and ask the AI to search for the image location worldwide. Alternatively, users can help the AI by narrowing down the search area by drawing a circle around the area of interest. The results will show the city, country, and GPS location with a confidence score indicating where the image was taken. Additionally, we display 100 possible locations on the map. We recommend users check all 100 results, especially if the confidence score from our AI is low.
API: Users can also use our API to predict image locations. The API is designed to integrate with various systems seamlessly. The API can return up to 100 results with confidence scores. Here is a simple example of how to use the Picarta API:
# First, pip install picarta
from picarta import Picarta
api_token = "YOUR_API_TOKEN"
localizer = Picarta(api_token)
# Geolocate a local image
result = localizer.localize(img_path="/path/to/local/image.jpg")
print(result)
More details about how to use our API can be found here.
Offline Model: Lastly, Picarta’s AI model can be installed on cloud or local servers for our clients. This option is available only for corporations and government entities. Please contact us to get an offer.
The Journey of Picarta
Our journey began with a simple yet ambitious idea: to create a tool that could identify the location of any photo taken anywhere in the world. In a digital age where the authenticity of visual content is often questioned, we saw an opportunity to make a difference. With the rise of fake news and the manipulation of images, verifying the origin of photos has become increasingly crucial. This is where Picarta steps in.
Join Us on Discord!
We’d love to hear from you and answer any questions you may have. Join our Discord community for the latest updates, discussions, and support directly from the Picarta team!