Grounding DINO Demo

Grounding DINO Demo

Finds objects in a picture from a text description, with no fixed list of labels. A Gradio demo of the Grounding DINO open-set detector, for developers and vision researchers.

Made by merve on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: This Space is a demo of Grounding DINO, an open-set object detector. Instead of choosing from a fixed list of classes, you describe what to find in words and the model looks for it in an image. The README presents it as the official PyTorch implementation and says it can run without a GPU.

Who it suits: It suits developers and researchers who want to try language-guided object detection, and people who need to locate things in pictures without training a detector first.

Industry | Who uses it | How they use it | Example
Computer vision research | Researchers | Try open-set detection with a plain text phrase | A researcher types a phrase and checks what the model finds
Software | Developers | Test language-guided detection before building it into a product | A developer runs the demo on a few sample images
Data labeling | Annotation teams | Locate objects by description to speed up labeling | A team tests whether a phrase finds the objects they need
Education | Students | See how a detector can follow words instead of fixed classes | A student tries different phrases on one picture

You can skip it if you need to train a detector, since the README lists the training code as not yet released.

Where to use it: You use it in these places.

• A web browser, on the Space's page on Hugging Face
• A local copy on Windows, macOS or Linux, started with app.py
• A command line script in the repository, which also has a CPU-only mode

Use case | Who | What they get back
Find objects by phrase | Anyone | Detections for the objects your phrase describes
Run without a GPU | Developers | The same demo in CPU-only mode
Script a detection | Developers | Results written to the output location you pass to the script

How to set it up: Online use needs only a browser. A local copy needs Python and the packages the repository lists.

• A web browser, for the online page
• Python with pip, for a local copy
• Python packages from requirements.txt
• CUDA_HOME set, if you have a CUDA environment. Without CUDA the code builds in CPU-only mode.

1. Open the Space's page on Hugging Face to use it online, with nothing to install.
2. For a local copy, install the Hugging Face Hub package.
3. Download the Space's files into a folder and move into it.
4. Install Gradio 3.23.0 and the packages in requirements.txt.
5. Start the app.

pip install -U huggingface_hub
hf download merve/Grounding_DINO_demo --repo-type space --local-dir Grounding_DINO_demo
cd Grounding_DINO_demo
pip install gradio==3.23.0
pip install -r requirements.txt
python app.py

The README also gives this install line for the project itself.

pip install -e .

How to use it: Open the Space's page, give it an image and a text phrase naming what to find, and read the detections. The README's own demo uses the phrase cat ear with a trailing period. The command line script takes a config, a checkpoint, an image, an output location and the phrase.

CUDA_VISIBLE_DEVICES=6 python demo/inference_on_a_image.py \
  -c /path/to/config \
  -p /path/to/checkpoint \
  -i .asset/cats.png \
  -o "outputs/0" \
  -t "cat ear." \
  [--cpu-only] # open it for cpu mode

Tool | What it does
-c | The path to the config file
-p | The path to the checkpoint file
-i | The input image
-o | Where the output goes
-t | The text phrase to look for
--cpu-only | Runs on the CPU, for machines without a GPU

Example: You pass a cat photo with the phrase cat ear. and the output location outputs/0 to the script. The model looks for the ears in the picture, and the results go to the output location.

For newcomers and experts: Most object detectors only know the classes they were trained on. This one reads a short phrase, so you can ask for something specific. Try the online page first with a simple phrase and one picture.

Under the hood, the README ties the model to DINO and GLIP and lists a Swin-T backbone for the released checkpoint. The record also names the BERT and RoBERTa text models. The README plans demos that pair it with Stable Diffusion and GLIGEN, but those are not yet released.

You give an image and a text phrase
The model matches the phrase to objects
Matching objects are returned
CPU-only mode runs without a GPU

Good to know: The README reports these detection scores, and its numbers differ between the highlight and the checkpoint listing.

Title: Detection scores reported on COCO
Highlight, zero-shot: 52.5 AP
Highlight, fine-tuned: 63.0 AP
GroundingDINO-T, zero-shot: 48.4 AP
GroundingDINO-T, fine-tuned: 57.2 AP

• The highlight says the zero-shot score came from training without COCO data.
• License: apache-2.
• Cost to run: none named. The Space uses cpu-basic hardware at Hugging Face.
• Data it reaches: the images you give it and the model files the record names, which come from Hugging Face.
• Not for: training a detector, since the README lists the training code as not yet released.

On Hugging Face, 13 November 2026

• Likes: 126
• Built with: gradio
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/merve/Grounding_DINO_demo
• Files: https://huggingface.co/spaces/merve/Grounding_DINO_demo/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.