Installation¶

The online documentation follows the repository's main branch. Choose the install that matches the API you want to use:

Source	Install	What you get
PyPI (stable)	`pip install torchcam`	Latest tagged release.
Git (`main`)	`pip install "torchcam @ git+https://github.com/frgfm/torch-cam.git"`	Unreleased changes matching this site.

The development version requires PyTorch 2.13 or higher within the 2.x series and supports both NumPy 1.26.4 and NumPy 2 (numpy>=1.26.4,<3). This floor includes fixes for GHSA-53q9-r3pm-6pq6, GHSA-63cw-57p8-fm3p, and GHSA-rrmf-rvhw-rf47. These version floors do not guarantee that future advisories will not affect a resolved environment. If you install torchvision for the demo or examples, use the release matching your PyTorch version; PyTorch 2.13 pairs with torchvision 0.28.

Matplotlib 3.8.4 or higher, below version 4, is also required. Earlier Matplotlib wheels can fail to import with NumPy 2.

Pillow 13 or higher is required for the core library and demo; its release notes list the image-parsing fixes.

On macOS, the PyTorch 2.13 wheels support Apple Silicon on macOS 14 or later. PyTorch does not publish matching Intel Mac wheels at this version floor.

Check the installed version when an example and your environment behave differently:

import importlib.metadata

print(importlib.metadata.version("torchcam"))

Virtual environment¶

Tip

You will need an environment manager, and I cannot recommend enough uv.

Create a virtual environment with your preferred Python version (3.11 or higher is required to use TorchCAM):

$ uv venv --python 3.11

StableLatest

$ uv pip install torchcam

$ uv pip install "torchcam @ git+https://github.com/frgfm/torch-cam.git"

System installation¶

You'll need Python 3.11 or higher, and a package installer like uv or pip.

StableLatestStable (pip)Latest (pip)

$ uv pip install --system torchcam

$ uv pip install --system "torchcam @ git+https://github.com/frgfm/torch-cam.git"

$ pip install torchcam

$ pip install "torchcam @ git+https://github.com/frgfm/torch-cam.git"

Info

TorchCAM is built on top of PyTorch which is a complex dependency. Proper installation depends on your system and available hardware. You can refer to installation guide of uv which is quite detailed.

For common pitfalls (e.g. no_grad with Grad-CAM, hook cleanup, picking target_layer), see Troubleshooting.