Spaces:
Sleeping
Sleeping
Create predict_utils.py
Browse files- predict_utils.py +78 -0
predict_utils.py
ADDED
|
@@ -0,0 +1,78 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import random
|
| 2 |
+
|
| 3 |
+
import torch
|
| 4 |
+
from PIL import Image
|
| 5 |
+
|
| 6 |
+
from config import IMAGE_SIZE
|
| 7 |
+
from data_utils import get_transform, load_charcoal_dataset
|
| 8 |
+
from train_utils import load_model, get_runtime_device
|
| 9 |
+
|
| 10 |
+
|
| 11 |
+
def predict_uploaded_image(model_name: str, image: Image.Image):
|
| 12 |
+
if not model_name:
|
| 13 |
+
return "Veuillez sélectionner un modèle.", None
|
| 14 |
+
|
| 15 |
+
if image is None:
|
| 16 |
+
return "Veuillez importer une image.", None
|
| 17 |
+
|
| 18 |
+
device = get_runtime_device()
|
| 19 |
+
model, meta = load_model(model_name, device)
|
| 20 |
+
|
| 21 |
+
class_names = meta["config"]["class_names"]
|
| 22 |
+
transform = get_transform()
|
| 23 |
+
|
| 24 |
+
image = image.convert("RGB")
|
| 25 |
+
tensor = transform(image).unsqueeze(0).to(device)
|
| 26 |
+
|
| 27 |
+
with torch.no_grad():
|
| 28 |
+
logits = model(tensor)
|
| 29 |
+
probs = torch.softmax(logits, dim=1).squeeze(0).cpu().tolist()
|
| 30 |
+
pred_idx = int(torch.argmax(logits, dim=1).item())
|
| 31 |
+
|
| 32 |
+
result_text = (
|
| 33 |
+
f"Prédiction : {class_names[pred_idx]}\n"
|
| 34 |
+
f"Confiance : {max(probs):.4f}\n\n"
|
| 35 |
+
f"Modèle : {model_name}\n"
|
| 36 |
+
f"Jeu de données : {meta['config']['dataset_name']}\n"
|
| 37 |
+
f"Appareil utilisé : {device}"
|
| 38 |
+
)
|
| 39 |
+
|
| 40 |
+
prob_dict = {class_names[i]: float(probs[i]) for i in range(len(class_names))}
|
| 41 |
+
return result_text, prob_dict
|
| 42 |
+
|
| 43 |
+
|
| 44 |
+
def test_random_sample(model_name: str):
|
| 45 |
+
if not model_name:
|
| 46 |
+
return None, "Veuillez sélectionner un modèle.", None
|
| 47 |
+
|
| 48 |
+
device = get_runtime_device()
|
| 49 |
+
model, meta = load_model(model_name, device)
|
| 50 |
+
|
| 51 |
+
raw, class_names = load_charcoal_dataset()
|
| 52 |
+
test_dataset = raw["test"]
|
| 53 |
+
|
| 54 |
+
idx = random.randint(0, len(test_dataset) - 1)
|
| 55 |
+
item = test_dataset[idx]
|
| 56 |
+
|
| 57 |
+
image = item["image"].convert("RGB")
|
| 58 |
+
label = int(item["label"])
|
| 59 |
+
label_name = class_names[label]
|
| 60 |
+
|
| 61 |
+
transform = get_transform()
|
| 62 |
+
tensor = transform(image).unsqueeze(0).to(device)
|
| 63 |
+
|
| 64 |
+
with torch.no_grad():
|
| 65 |
+
logits = model(tensor)
|
| 66 |
+
probs = torch.softmax(logits, dim=1).squeeze(0).cpu().tolist()
|
| 67 |
+
pred_idx = int(torch.argmax(logits, dim=1).item())
|
| 68 |
+
|
| 69 |
+
result_text = (
|
| 70 |
+
f"Échantillon test aléatoire\n"
|
| 71 |
+
f"Vérité terrain : {label_name}\n"
|
| 72 |
+
f"Prédiction : {class_names[pred_idx]}\n"
|
| 73 |
+
f"Confiance : {max(probs):.4f}\n"
|
| 74 |
+
f"Appareil utilisé : {device}"
|
| 75 |
+
)
|
| 76 |
+
|
| 77 |
+
prob_dict = {class_names[i]: float(probs[i]) for i in range(len(class_names))}
|
| 78 |
+
return image, result_text, prob_dict
|