vit-base-blur

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

• Loss: 0.0008
• Accuracy: 1.0

Model description

Model trained for binary classification between 'noisy' (blurry) and clean images, where 'noisy' images are the result of unfinished/insufficient passes from an LDM for image generation

Intended uses & limitations

More information needed

Training and evaluation data

1000ish clean and blurry images using 30 and 10 steps respectively on SD2.1

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0001
• 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: 12

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.0082	1.02	100	0.0107	1.0
0.0079	2.04	200	0.0052	1.0
0.0029	3.06	300	0.0028	1.0
0.002	4.08	400	0.0020	1.0
0.0016	5.1	500	0.0015	1.0
0.0013	6.12	600	0.0013	1.0
0.0011	7.14	700	0.0011	1.0
0.001	8.16	800	0.0010	1.0
0.0009	9.18	900	0.0009	1.0
0.0008	10.2	1000	0.0008	1.0
0.0008	11.22	1100	0.0008	1.0

Framework versions

• Transformers 4.30.2
• Pytorch 2.0.1+cu118
• Datasets 2.13.1
• Tokenizers 0.13.3