emotion_classification_model

This model is a fine-tuned version of google/vit-base-patch16-224-in21k on the imagefolder dataset. It achieves the following results on the evaluation set:

• Loss: 1.5450
• Accuracy: 0.525

Model description

This is an image classification model fine-tuned from a pre-trained Vision Transformer (ViT). It's designed to classify emotions from human facial images by analyzing visual expressions.

Intended uses & limitations

Intended uses This model is primarily for image-based emotion classification. Potential applications include sentiment analysis, human-computer interaction (HCI), and psychological research.

Limitations Current limitations include its limited accuracy (0.525), potential biases from training data, challenges with subtle or ambiguous emotional expressions, and sensitivity to image quality. Privacy concerns should also be considered when deploying this model.

Training and evaluation data

The model was trained and evaluated on an imagefolder dataset, which is assumed to contain various images categorized by emotion (e.g., happy, sad, angry). The reported accuracy of 0.525 is based on its performance on the evaluation set of this dataset.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 30

Training results

Framework versions

• Transformers 4.53.1
• Pytorch 2.7.1+cpu
• Datasets 4.0.0
• Tokenizers 0.21.2