font-identifier

This model is a fine-tuned version of microsoft/resnet-18 on the imagefolder dataset. Result: Loss: 0.1172; Accuracy: 0.9633

Try with any screenshot of a font, or any of the examples in the 'samples' subfolder of this repo.

Model description

Identify the font used in an image. Visual classifier based on ResNet18.

I built this project in 1 day, with a minute-by-minute journal on Twitter/X, on Pebble.social, and on Threads.net.

The code used to build this model is in this github rep

Intended uses & limitations

Identify any of 48 standard fonts from the training data.

Training and evaluation data

Trained and eval'd on the gaborcselle/font-examples dataset (80/20 split).

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
4.0243	0.98	30	3.9884	0.0204
0.8309	10.99	338	0.5536	0.8551
0.3917	20.0	615	0.2353	0.9388
0.2298	30.99	953	0.1326	0.9633
0.1804	40.0	1230	0.1421	0.9571
0.1987	46.99	1445	0.1250	0.9673
0.1728	48.0	1476	0.1293	0.9633
0.1337	48.78	1500	0.1172	0.9633

Confusion Matrix

Confusion matrix on test data.

Framework versions

• Transformers 4.36.0.dev0
• Pytorch 2.0.0
• Datasets 2.12.0
• Tokenizers 0.14.1