DINOHash Perceptual Hash

This repository provides functionality for generating perceptual hashes for images using DinoHash

NOTE: we suggest using git clone ... --depth 1 while setting up the repository since there were large files in previous commits that might cause git to malfunction

Node.js Implementation

To use the Node.js package look at the README in the node_package folder

Setup

Install the latest version of PyTorch according to your OS. Then run

pip install transformers numpy pillow

Usage

dinohash Function

The dinohash function generates perceptual hashes for the given images using the DINOv2 model.

Parameters

• image_arrays (Union[np.ndarray, List[Image.Image], torch.Tensor]): Input images as a numpy array, a list of PIL Images, or a torch.Tensor.

Returns

• List[Hash]: The generated perceptual hashes.

Example

from PIL import Image
import torch
from dinohash import DINOHash

# Load an image
image_path = "path/to/your/image.jpg"
image = Image.open(image_path)

# Create DINOHash instance
dinohash = DINOHash()

# Generate perceptual hash
hashes = dinohash.hash([image])

# Print the hexadecimal representation of the hash
print("Perceptual hash:", hashes[0].hex)

Command Line Usage

You can also use the dinohash.py script from the command line to generate perceptual hashes for an image.

python dinohash.py <image_path>

Replace <image_path> with the path to your image file.

Example:

python dinohash.py path/to/your/image.jpg

This will print the perceptual hash of the image.