Calfin

Calfin is a semantic image segmentation model available on Hugging Face. It is implemented in PyTorch and utilizes the segmentation-models-pytorch library. The model is based on a Feature Pyramid Network (FPN) architecture, with a ResNet34 encoder pretrained on ImageNet. Key initialization parameters for Calfin include an encoder depth of 5, 256 pyramid channels in the decoder, 128 segmentation channels in the decoder, a merge policy set to 'add', a decoder dropout rate of 0.2, nearest neighbor interpolation in the decoder, three input channels, one output class, no activation function specified, an upsampling factor of four, and no auxiliary parameters. The model is trained and evaluated on the CALFIN dataset, as indicated in the documentation. Performance metrics reported for Calfin include a test per-image intersection-over-union (IoU) of approximately 0.54 and a test dataset IoU of approximately 0.52. The model can be loaded using the segmentation-models-pytorch library, with pretrained weights accessible from the repository. Calfin is released under the MIT license. No additional information is provided about intended users, deployment platforms beyond PyTorch, or specific application domains. The tool is positioned as a model for semantic segmentation tasks, particularly with the CALFIN dataset.