Cracks_dataset_2026_convnext_model

This model is a fine-tuned version of facebook/convnext-base-224-22k on the Logiroad/Cracks_dataset_2026_splits1000 dataset. It achieves the following results on the evaluation set:

• Loss: 0.7788
• Accuracy: 0.7413

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: 1e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 1337
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.05
• num_epochs: 30.0
• mixed_precision_training: Native AMP
• label_smoothing_factor: 0.05

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
1.3158	1.0	302	1.2774	0.4030
1.0948	2.0	604	1.0307	0.6045
1.0577	3.0	906	0.9374	0.6480
0.9709	4.0	1208	0.8760	0.6766
0.8839	5.0	1510	0.8935	0.6654
0.8254	6.0	1812	0.8503	0.6990
0.9047	7.0	2114	0.8020	0.7114
0.9098	8.0	2416	0.8173	0.7127
0.8357	9.0	2718	0.8086	0.7002
0.8455	10.0	3020	0.8114	0.6828
0.8015	11.0	3322	0.7899	0.7164
0.8386	12.0	3624	0.8763	0.6803
0.7725	13.0	3926	0.7981	0.7065
0.7663	14.0	4228	0.7917	0.7264
0.7314	15.0	4530	0.8059	0.7127
0.7152	16.0	4832	0.8065	0.7127
0.6853	17.0	5134	0.7929	0.7251
0.6986	18.0	5436	0.7788	0.7413
0.6992	19.0	5738	0.7908	0.7226
0.6889	20.0	6040	0.8038	0.7276
0.6616	21.0	6342	0.8265	0.7114
0.6589	22.0	6644	0.8011	0.7301
0.7379	23.0	6946	0.8198	0.7090
0.6684	24.0	7248	0.7985	0.7264
0.6984	25.0	7550	0.7998	0.7214
0.6881	26.0	7852	0.7948	0.7313
0.6624	27.0	8154	0.8071	0.7102
0.6727	28.0	8456	0.8075	0.7152
0.7003	29.0	8758	0.8049	0.7201
0.6273	30.0	9060	0.8048	0.7177

Framework versions

• Transformers 4.46.1
• Pytorch 2.3.0
• Datasets 3.1.0
• Tokenizers 0.20.3