Poke-Grader Grade Regressor (efficientnet_b4)
Fine-tune de efficientnet_b4 para estimar la nota PSA (1–10) de cartas Pokémon TCG.
Salidas del modelo
| Salida | Descripción |
|---|---|
overall_grade |
Nota global PSA (1–10) |
surface |
Estado de la superficie |
corners |
Estado de las esquinas |
edges |
Estado de los bordes |
centering |
Centrado de la imagen |
Métricas (mejor epoch: 17)
| Métrica | Valor |
|---|---|
| val_loss | 0.0018 |
| MAE (escala PSA 1-10) | ±0.399 puntos |
Entrenado con 37 épocas (early stopping patience=12).
Uso (ONNX)
import onnxruntime as ort
import numpy as np
sess = ort.InferenceSession('card_grade_regressor.onnx')
# img: np.array float32, shape (1, 3, 380, 380), normalizado ImageNet
# Output en [0, 1] → reescalar a [1, 10]: grade = output * 9 + 1
grades = sess.run(None, {'image': img})[0]
grades_psa = grades * 9 + 1 # [overall, surface, corners, edges, centering]
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Evaluation results
- Mean Absolute Error (MAE)self-reported0.399