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

Loss: 1.9156
Accuracy: 0.5852

Model description

First generation of my AI that tells you what car you took a picture of.
More versions coming soon with accuracy ratings of 85% and higher! Trained on 70+ brands with 2700+ cars going from 1945-2024.
App coming soon (also called Caracam) to Android and IOS
(Late March - Early April 2024). In the future I will take user opinion into account on what brands to add. The app will be updated semi-yearly with user-suggested car brands!
if you wish to support project Caracam please visit my Patreon or my Cashapp!!

Intended uses & limitations

NOT FOR COMMERCIAL USE OUTSIDE OF OFFICIAL CARACAM MOBILE APP

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 5e-05
train_batch_size: 32
eval_batch_size: 32
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 128
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_ratio: 0.1
num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
4.0308	1.0	5362	3.6948	0.2491
2.694	2.0	10725	2.2586	0.5199
2.4475	3.0	16086	1.9156	0.5852

Framework versions

Transformers 4.36.2
Pytorch 2.1.2+cpu
Datasets 2.16.1
Tokenizers 0.15.0

Takekazuchi/Caracam

Caracam (gen 1)