Code used in fungi image analysis, supervised and unsupervised. Effort described in Towards Data Science, see https://towardsdatascience.com/image-clustering-implementation-with-pytorch-587af1d14123 (no paywall).
The fungi image data is loaded and pre-procssed in fungidata.py
in which the DataSet class is created through a factory method. That includes full images, grid images, with or without ground-truth label or index in dataset. The specific image dataset is presently proprietary, but can be recreated from the Danish fungi atlas, see https://svampe.databasen.org
Image classification efforts are in files starting with ic
. The template models for example are loaded in ic_template_models.py
. The auto-encoder is defined in ae_deep.py
with a learner class in ae_learner
. Local Aggregation criterion is found in cluster_utils
and its learner class in la_learner
. The training inherits from _learner
. To use the implementation for another custom dataset, modify how self.dataset
is set in _learner
.