FacialConfidence

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

vit-Facial-Confidence

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

• Loss: 0.2560
• Accuracy: 0.8970

Model description

Facial Confidence is an image classification model which takes a black and white image of a persons headshot and classifies it as confident or unconfident.

Intended uses & limitations

The model is intended to help with behavioral analysis tasks. The model is limited to black and white images where the image is a zoomed in headshot of a person (For best output the input image should be as zoomed in on the subjects face as possible without cutting any aspects of their head)

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0002
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 4
• mixed_precision_training: Native AMP

Training results

Framework versions

• Transformers 4.44.2
• Pytorch 2.5.0+cu121
• Datasets 3.0.2
• Tokenizers 0.19.1