license: apache-2.0
base_model: thezeivier/Grietas_10k
tags:
- generated_from_trainer
metrics:
- accuracy
model-index:
- name: Grietas_10k-Fine-tuning
results: []
Grietas_10k-Fine-tuning
This model is a fine-tuned version of thezeivier/Grietas_10k on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.3864
- Accuracy: 0.8860
Model description
More information needed
Intended uses & limitations
Este modelo ha sido diseñado para la clasificación de imágenes de infraestructuras en tres categorías:
- Sano (sin daños en la estructura de concreto).
- Fisura (daños leves e insignificantes en la estructura de concreto).
- Grieta (daños graves y de alto riesgo en la estructura de concreto).
Este modelo de visión artificial puede ser una herramienta valiosa para identificar posibles amenazas de colapso en estructuras de concreto en caso de futuros terremotos.
Limitaciones: El modelo se ha entrenado exclusivamente con imágenes correspondientes a las tres categorías mencionadas anteriormente y no incorpora información sobre la distancia entre la cámara y la grieta capturada en la imagen.
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
Los siguientes hiperparámetros fueron utilizados durante el entrenamiento:
- learning_rate (tasa de aprendizaje): 5e-05
- train_batch_size (tamaño del lote de entrenamiento): 80
- eval_batch_size (tamaño del lote de evaluación): 32
- seed (semilla): 42
- gradient_accumulation_steps (pasos de acumulación de gradientes): 4
- total_train_batch_size (tamaño total del lote de entrenamiento): 320
- optimizer (optimizador): Adam con betas=(0.9,0.999) y epsilon=1e-08
- lr_scheduler_type (tipo de programador de tasa de aprendizaje): lineal
- lr_scheduler_warmup_ratio (proporción de calentamiento del programador de tasa de aprendizaje): 0.1
- num_epochs (número de épocas): 100
Estos hiperparámetros fueron utilizados para entrenar el modelo y pueden ser configurados en la parte correspondiente del modelo para replicar las mismas condiciones de entrenamiento. Cada hiperparámetro tiene un impacto en cómo se ajusta el modelo a los datos y puede afectar su rendimiento y velocidad de entrenamiento, por lo que es importante seleccionarlos cuidadosamente.
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|---|---|---|---|---|
| No log | 0.8 | 2 | 1.3737 | 0.3679 |
| No log | 2.0 | 5 | 1.0234 | 0.6218 |
| No log | 2.8 | 7 | 0.8146 | 0.7254 |
| 1.0488 | 4.0 | 10 | 0.6621 | 0.7772 |
| 1.0488 | 4.8 | 12 | 0.6295 | 0.8031 |
| 1.0488 | 6.0 | 15 | 0.5390 | 0.8083 |
| 1.0488 | 6.8 | 17 | 0.4902 | 0.8290 |
| 0.4981 | 8.0 | 20 | 0.4645 | 0.8290 |
| 0.4981 | 8.8 | 22 | 0.4484 | 0.8497 |
| 0.4981 | 10.0 | 25 | 0.4543 | 0.8446 |
| 0.4981 | 10.8 | 27 | 0.4325 | 0.8394 |
| 0.3669 | 12.0 | 30 | 0.4210 | 0.8497 |
| 0.3669 | 12.8 | 32 | 0.4303 | 0.8342 |
| 0.3669 | 14.0 | 35 | 0.4170 | 0.8497 |
| 0.3669 | 14.8 | 37 | 0.3861 | 0.8601 |
| 0.2811 | 16.0 | 40 | 0.3629 | 0.8705 |
| 0.2811 | 16.8 | 42 | 0.3982 | 0.8653 |
| 0.2811 | 18.0 | 45 | 0.4492 | 0.8290 |
| 0.2811 | 18.8 | 47 | 0.4216 | 0.8342 |
| 0.2026 | 20.0 | 50 | 0.4614 | 0.8394 |
| 0.2026 | 20.8 | 52 | 0.4325 | 0.8446 |
| 0.2026 | 22.0 | 55 | 0.4755 | 0.8342 |
| 0.2026 | 22.8 | 57 | 0.4175 | 0.8394 |
| 0.1709 | 24.0 | 60 | 0.4175 | 0.8497 |
| 0.1709 | 24.8 | 62 | 0.4105 | 0.8446 |
| 0.1709 | 26.0 | 65 | 0.4140 | 0.8601 |
| 0.1709 | 26.8 | 67 | 0.4641 | 0.8394 |
| 0.1293 | 28.0 | 70 | 0.4214 | 0.8394 |
| 0.1293 | 28.8 | 72 | 0.3802 | 0.8808 |
| 0.1293 | 30.0 | 75 | 0.4875 | 0.8290 |
| 0.1293 | 30.8 | 77 | 0.3972 | 0.8705 |
| 0.1167 | 32.0 | 80 | 0.4853 | 0.8394 |
| 0.1167 | 32.8 | 82 | 0.4082 | 0.8549 |
| 0.1167 | 34.0 | 85 | 0.3917 | 0.8601 |
| 0.1167 | 34.8 | 87 | 0.3573 | 0.8653 |
| 0.1034 | 36.0 | 90 | 0.4312 | 0.8497 |
| 0.1034 | 36.8 | 92 | 0.4035 | 0.8497 |
| 0.1034 | 38.0 | 95 | 0.4413 | 0.8238 |
| 0.1034 | 38.8 | 97 | 0.4728 | 0.8446 |
| 0.0782 | 40.0 | 100 | 0.3977 | 0.8808 |
| 0.0782 | 40.8 | 102 | 0.3449 | 0.8912 |
| 0.0782 | 42.0 | 105 | 0.4146 | 0.8808 |
| 0.0782 | 42.8 | 107 | 0.4380 | 0.8601 |
| 0.083 | 44.0 | 110 | 0.4579 | 0.8497 |
| 0.083 | 44.8 | 112 | 0.5234 | 0.8549 |
| 0.083 | 46.0 | 115 | 0.4053 | 0.8756 |
| 0.083 | 46.8 | 117 | 0.4724 | 0.8394 |
| 0.0741 | 48.0 | 120 | 0.4631 | 0.8549 |
| 0.0741 | 48.8 | 122 | 0.4351 | 0.8653 |
| 0.0741 | 50.0 | 125 | 0.4191 | 0.8756 |
| 0.0741 | 50.8 | 127 | 0.3772 | 0.8964 |
| 0.067 | 52.0 | 130 | 0.3960 | 0.8808 |
| 0.067 | 52.8 | 132 | 0.3749 | 0.8964 |
| 0.067 | 54.0 | 135 | 0.4395 | 0.8653 |
| 0.067 | 54.8 | 137 | 0.5284 | 0.8342 |
| 0.0632 | 56.0 | 140 | 0.3332 | 0.8808 |
| 0.0632 | 56.8 | 142 | 0.4342 | 0.8497 |
| 0.0632 | 58.0 | 145 | 0.3986 | 0.8756 |
| 0.0632 | 58.8 | 147 | 0.4771 | 0.8549 |
| 0.063 | 60.0 | 150 | 0.4505 | 0.8497 |
| 0.063 | 60.8 | 152 | 0.4023 | 0.8653 |
| 0.063 | 62.0 | 155 | 0.5208 | 0.8290 |
| 0.063 | 62.8 | 157 | 0.4915 | 0.8601 |
| 0.0571 | 64.0 | 160 | 0.4412 | 0.8756 |
| 0.0571 | 64.8 | 162 | 0.4554 | 0.8653 |
| 0.0571 | 66.0 | 165 | 0.4318 | 0.8653 |
| 0.0571 | 66.8 | 167 | 0.4317 | 0.8549 |
| 0.0608 | 68.0 | 170 | 0.4509 | 0.8653 |
| 0.0608 | 68.8 | 172 | 0.4176 | 0.8705 |
| 0.0608 | 70.0 | 175 | 0.5203 | 0.8394 |
| 0.0608 | 70.8 | 177 | 0.4375 | 0.8756 |
| 0.0478 | 72.0 | 180 | 0.4196 | 0.8601 |
| 0.0478 | 72.8 | 182 | 0.4744 | 0.8601 |
| 0.0478 | 74.0 | 185 | 0.4362 | 0.8808 |
| 0.0478 | 74.8 | 187 | 0.4804 | 0.8653 |
| 0.0519 | 76.0 | 190 | 0.4861 | 0.8446 |
| 0.0519 | 76.8 | 192 | 0.4605 | 0.8601 |
| 0.0519 | 78.0 | 195 | 0.4730 | 0.8394 |
| 0.0519 | 78.8 | 197 | 0.4650 | 0.8705 |
| 0.0553 | 80.0 | 200 | 0.3864 | 0.8860 |
Framework versions
- Transformers 4.33.1
- Pytorch 2.0.1+cu118
- Datasets 2.14.5
- Tokenizers 0.13.3