OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Count Anything

Count Anything is an open-source Python computer-vision model that counts objects in an image from a category name or natural-language prompt and returns point-based instance locations.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Count Anything — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use the model when a counting task can be expressed as a text query over an image, or when you need code and a benchmark-oriented research workflow rather than only a hosted demo.

Who it is for

  • Computer-vision researchers studying text-guided object counting
  • Developers exploring image-counting or visual inventory workflows
  • Teams evaluating counting across varied image domains

Documented uses

  • Count a prompted object category in a single image.
  • Inspect predicted instance locations alongside the total count.
  • Prepare CLOC-based training, validation or test workflows.
Before you choose

Requirements & limits

The repository is public and lists an Apache-2.0 license. A linked Hugging Face demo is presented as a separate way to try the model.

  • Single-image inference requires the released Count Anything checkpoint at checkpoints/count_anything.pt.
  • Training or fine-tuning from SAM3 initialization additionally requires the official SAM3 pretrained weights, which the repository does not redistribute.
  • The repository uses CLOC by default for training and evaluation; dataset preparation is required for those workflows.
  • Predictions include points and a count; saved outputs also include point positions and confidence scores in JSON.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Set up single-image inference

Prepare the documented Python environment and checkpoint for the repository’s single-image inference example.

Before you start

  • Conda
  • Python 3.12
  • Repository dependencies from requirements.txt
  • Released Count Anything checkpoint, count_anything.pt

Steps

  1. Create and activate the countanything conda environment with Python 3.12.

  2. Install the repository dependencies with pip install -r requirements.txt.

  3. Download count_anything.pt and place it at checkpoints/count_anything.pt.

Expected outcome

The documented checkpoint path and environment are prepared for the single-image inference example, which takes an image path and text query and saves a point-and-count visualization and prediction JSON.

Guide limitations

  • The setup steps alone do not run inference; use the repository’s Single-image Inference instructions for the model call.
  • If pip selects an unsuitable CUDA build, the README says to install PyTorch and torchvision builds matching the local CUDA version.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
14 October 2026
Original website title
Mengqi-Lei/count-anything · Official homepage title · retained in its source language
How we review sourcesSuggest a correction