environment-classifier.yml

name: cameratraps-classifier

channels:
- pytorch
- defaults
- conda-forge

dependencies:
- python>=3.7, <=3.7.6

# from pytorch channel
- pytorch==1.9.0
- torchvision==0.10.0
# - cpuonly  # for CPU only
- cudatoolkit=11.3  # for GPU

# from defaults channel
- humanfriendly
- ipywidgets    # for using tqdm inside Jupyter Lab
- jsonpickle    # used by run_tf_detector_batch.py
- jupyterlab
- matplotlib
- mypy
- nb_conda_kernels
- networkx      # used for taxonomy graph
- numpy
- openpyxl      # used by taxonomy_mapping/species_by_dataset.py
- pandas
- progressbar2  # used by ai4eutils (which is called by taxonomy-mapping code)
- pylint
- python-graphviz  # used for taxonomy visualization
- scikit-learn
- seaborn>=0.11
- tqdm
- xlrd          # for pandas Excel support, needed by taxonomy_mapping/process_species_by_dataset
- pip

# from conda-forge
- accimage      # faster Pillow replacement for torchvision

# Some packages or versions are only available from PyPI
# These may require GCC to be installed
- pip:
#  - azure-cosmos
  - azure-storage-blob>=12.5  # require v12.5.0 for BlobClient.exists()

  # Pillow-SIMD will "over-shadow" the normal Pillow package installed by conda.
  # After installing, check that Pillow-SIMD is what gets imported:
  #     import PIL
  #     print(PIL.__version__)  # make sure this ends in '.postX'
  - pillow-simd

  # conda (defaults channel) tends to lag PyPI for TensorFlow-related packages
  - tensorboard
#  - tensorflow>=2.3  # require v2.3.0 or greater for Keras EfficientNet