The goal of imfeatures is to …

Install the development version from GitHub with:

# install.packages("devtools")
devtools::install_github("bbuchsbaum/imfeatures")

This is a basic example which shows you how to solve a common problem:

library(imfeatures)
## basic example code
im1 = "testdata/1_A_1.jpeg"
im2 = "testdata/1_B_1.jpeg"

im_feature_sim(c(im1,im2), layers=c(1,2,3))
#> [========================================================================================] 100%
#> $layer_1
#>            1_A_1.jpeg 1_B_1.jpeg
#> 1_A_1.jpeg  0.0000000  0.3326229
#> 1_B_1.jpeg  0.3326229  0.0000000
#> 
#> $layer_2
#>            1_A_1.jpeg 1_B_1.jpeg
#> 1_A_1.jpeg    0.00000    0.37037
#> 1_B_1.jpeg    0.37037    0.00000
#> 
#> $layer_3
#>            1_A_1.jpeg 1_B_1.jpeg
#> 1_A_1.jpeg  0.0000000  0.4695186
#> 1_B_1.jpeg  0.4695186  0.0000000

On HPC systems, avoid automatic Conda setup and use an existing Python instead.

1. Disable auto Python setup during load (recommended on HPC):

Sys.setenv(IMFEATURES_SKIP_PYTHON = "TRUE")

To make this permanent, add IMFEATURES_SKIP_PYTHON=TRUE to ~/.Renviron.

2. Create or choose a Python environment (module + venv is typical):

# Example using your cluster's Python module
module load python/3.10             # if applicable (3.9/3.10 recommended)
python -m venv $WORK/venvs/imfeatures
source $WORK/venvs/imfeatures/bin/activate
pip install --upgrade pip wheel setuptools

# Minimal required packages
pip install Pillow numpy

# Optional packages for full functionality
pip install thingsvision resmem open-clip-torch

# PyTorch: use your cluster's module if available, otherwise pick one index
# CPU only:
pip install torch torchvision --index-url https://download.pytorch.org/whl/cpu
# or CUDA-specific (example):
# pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118

Note on Python versions: some optional packages (e.g., thingsvision via numba)
currently do not support Python 3.11+. Prefer Python 3.9 or 3.10 when you plan
to use these features.

3. Tell imfeatures to use that Python:

library(imfeatures)
use_existing_python("$WORK/venvs/imfeatures/bin/python")

Alternatively, set once in ~/.Renviron and it will be auto-detected:

RETICULATE_PYTHON=$WORK/venvs/imfeatures/bin/python
IMFEATURES_SKIP_PYTHON=TRUE

4. Verify:

reticulate::py_config()

Troubleshooting:

• If you see a Conda error like “bad interpreter” during library(imfeatures), it’s usually a broken R-miniconda on shared filesystems. Use the HPC Quickstart above or set IMFEATURES_SKIP_PYTHON=TRUE and call use_existing_python().
• To force virtualenv instead of Conda during imfeatures_config(), set IMFEATURES_METHOD=virtualenv before calling it.