Swin T Food101

Swin T Food101 is a model checkpoint for image classification tasks, available on Hugging Face. The model is based on the Swin Transformer architecture and has been trained using the Food-101 dataset, which is referenced as ethz/food101. The checkpoint file, named Swin-T-food101.pt, is a raw export from a training run, and accompanying configuration files from Hydra are also provided. Performance metrics from the training process are included: on the training split, the model achieved a loss of 0.0139, with top-1 accuracy of 99.63% and top-5 accuracy of 99.97%. On the validation split, the loss was 1.6996, with top-1 accuracy at 69.97% and top-5 accuracy at 88.88%. The model is distributed under the MIT license. It is implemented in PyTorch, as indicated by the tags and file formats. The files provided are intended for users who are able to work with raw model checkpoints, as the evidence notes that this is a raw checkpoint dump. There is no mention of a hosted inference API or deployment by an inference provider, so users must download and use the checkpoint in their own environments. The evidence does not specify a particular target audience, but the technical nature of the files and the training details suggest it is suitable for those with experience in machine learning or computer vision who wish to work with a Swin Transformer model trained on food image classification. No information is given about integrations, additional features, or specific use cases beyond image classification on the Food-101 dataset. Pricing details are not mentioned, but the MIT license indicates it is open source. The model card does not provide further context about its intended applications or broader positioning within the field.