GeCo2 Object Counting

GeCo2 Object Counting

Counts all instances of an object in an image after you mark a few examples, with no retraining or fixed list of classes. Suits researchers and anyone counting items in photos.

Made by jerpelhan 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: GeCo2 counts objects in a picture. You mark a few examples of the thing you care about, and it finds and counts every matching instance in the image. It needs no fixed list of object classes and no retraining. This Space is a Gradio demo of the GeCo2 model, and it is based on an earlier counting demo called CountGD.

Who it suits: It suits researchers working on few-shot counting, and people who need to count unusual objects where labelled data is scarce.

Industry | Who uses it | How they use it | Example
Retail and stock | Inventory staff | Count items shown in a photo | Mark one box on a shelf and count the rest
Agriculture | Growers and agronomists | Count plants or fruit in an image | Mark two fruits and count the others in the tree photo
Lab work | Technicians | Count small objects when no trained counter exists | Mark a few samples and count all similar ones in a picture
Research | Computer vision researchers | Try few-shot counting on new object types | Compare counts from different example choices

Skip it if you want a counter that needs no examples, since this one asks you to mark a few.

Where to use it:

• A browser: open the Space's page on Hugging Face, with nothing to install
• A copy on Windows, macOS or Linux, using Python 3.10.13

Use case | Who | What they get back
Counting objects in a photo | Anyone with a picture | All matching instances found and counted
Exploring a new category | Researchers | A quick count without training a new model
Reading the code | Developers | A repository with SAM 2 and Deformable-DETR folders

How to set it up: The online copy needs nothing. For a copy on your own computer, you need the following.

• Python 3.10.13 with pip on Windows, macOS or Linux
• The Hugging Face hub client, which the first step installs
• The packages listed in requirements.txt

1. Install the Hugging Face hub client.
2. Download the Space's files into a local folder.
3. Move into that folder.
4. Install the Gradio version the Space pins.
5. Install the remaining packages from requirements.txt.
6. Start demo_gradio.py.

pip install -U huggingface_hub
hf download jerpelhan/GECO2-demo --repo-type space --local-dir GECO2-demo
cd GECO2-demo
pip install gradio==5.50.0
pip install -r requirements.txt
python demo_gradio.py

How to use it: Upload an image, then mark a few examples of the object you want counted. GeCo2 spreads that idea across the picture and counts the matches. The source does not describe the exact controls, so follow the labels on the page.

python demo_gradio.py

Example: You upload a photo of a bowl of oranges and mark two or three of them as examples. The Space returns the photo with the matching oranges detected, so you can read off how many there are. The source gives no accuracy figures.

For newcomers and experts: If you are new, pick a picture with many similar objects in clear view and mark two or three clean examples. Add one more example if the first count looks wrong.

If you know the field, GeCo2 detects and counts by learning the target from a handful of exemplars, with no fixed class list, then applying it over the whole image. The Space names four SAM 2 Hiera sizes, from tiny to large, and runs on zero-a10g hardware. The paper's claims on scale variation and varied domains are the author's own.

You upload an image
You mark a few example objects
GeCo2 spreads the concept over the image
Matching objects are detected and counted

Good to know: The source is a demo page, so speed and accuracy limits are not described.

• License: mit, as listed on Hugging Face
• Cost to run: no API key or paid service is named, and the online copy runs on zero-a10g hardware at Hugging Face
• Data it reaches: the pictures you upload, which the online Space handles on Hugging Face's servers, plus an assets dataset the Space names
• Not for: the source lists no exclusions, so use only pictures you have the right to use

On Hugging Face, 13 November 2026

• Likes: 11
• Built with: gradio
• License: mit

Links

• Space: https://huggingface.co/spaces/jerpelhan/GECO2-demo
• Files: https://huggingface.co/spaces/jerpelhan/GECO2-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.