vit-base-avengers-v1

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: 0.5324
• Accuracy: 0.8683

Refer to this medium article for more info on how it was trained.

Limitations

Training was done on google images for these search terms each representing a class. Iron Man,Captain America,Thor,Spider Man,Docter Strage,Black Panther,Ant Man,Captain Marvel,Hulk,Black Widow,Hawkeye Avengers,Scarlet Witch,Vision Avengers,Bucky Barnes,Falcon Avengers,Loki

Therefore it has seen more of images where these super heros are in their suit or superhero outfit. For example an image of hulk is detected correctly, but an image of Bruce Banner is not simply because the model has't seen those images. A little bit of data augmentation will help.

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

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

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.8183	1.27	100	1.0134	0.8464
0.2234	2.53	200	0.6146	0.8495
0.1206	3.8	300	0.5324	0.8683

Framework versions

• Transformers 4.20.1
• Pytorch 1.11.0+cu113
• Datasets 2.3.2
• Tokenizers 0.12.1