MedicalVision/test_remove

MedicalVision

Detección de objetos

Modelo de detección de objetos utilizando transformadores, basado en yolos y safetensors. Este modelo se despliega a través de Endpoints de Inference. El rendimiento del modelo mejora después del entrenamiento, especialmente en métricas clave como Precision Media y Recall Medio.

Como usar

Para utilizar este modelo, puedes seguir estas métricas de evaluación para ver su rendimiento:
Original result

* IoU metric: bbox
* Precisión Media (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.005
* Precisión Media (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.005
* Precisión Media (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.005
* Precisión Media (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.203
* Precisión Media (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.068
* Precisión Media (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.005
* Recall Medio (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.029
* Recall Medio (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.029
* Recall Medio (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.029
* Recall Medio (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.200
* Recall Medio (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.067
* Recall Medio (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.029

After training result

* IoU metric: bbox
* Precisión Media (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.009
* Precisión Media (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.020
* Precisión Media (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.008
* Precisión Media (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.000
* Precisión Media (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.000
* Precisión Media (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.009
* Recall Medio (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.043
* Recall Medio (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.076
* Recall Medio (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.087
* Recall Medio (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.000
* Recall Medio (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.000
* Recall Medio (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.089

Config

* dataset: VinXray
* modelo original: hustvl/yolos-tiny
* lr: 0.0001
* dropout_rate: 0.1
* weight_decay: 0.0001
* max_epochs: 1
* train samples: 67234

Logging

* validation_loss: tensor(8.5927, device='cuda:0')
* validation_loss_ce: tensor(3.4775, device='cuda:0')
* validation_loss_bbox: tensor(0.5756, device='cuda:0')
* validation_loss_giou: tensor(1.1184, device='cuda:0')
* validation_cardinality_error: tensor(99.5938, device='cuda:0')
* training_loss: tensor(1.3630, device='cuda:0')
* train_loss_ce: tensor(0.2593, device='cuda:0')
* train_loss_bbox: tensor(0.0803, device='cuda:0')
* train_loss_giou: tensor(0.3511, device='cuda:0')
* train_cardinality_error: tensor(0.5294, device='cuda:0')
* validation_loss: tensor(1.5262, device='cuda:0')
* validation_loss_ce: tensor(0.2351, device='cuda:0')
* validation_loss_bbox: tensor(0.0827, device='cuda:0')
* validation_loss_giou: tensor(0.4389, device='cuda:0')
* validation_cardinality_error: tensor(0.4794, device='cuda:0')

Ejemplos

* size: tensor([560, 512])
* image_id: tensor([1])
* class_labels: tensor([], dtype=torch.int64)
* boxes: tensor([], size=(0, 4))
* area: tensor([])
* iscrowd: tensor([], dtype=torch.int64)
* orig_size: tensor([2580, 2332])

Funcionalidades

Detección de objetos
Transformadores
Compatibilidad con Safetensors

Casos de uso

Detección de objetos en imágenes médicas