FIX

app.py

@@ -3,23 +3,23 @@ from transformers import pipeline
 from PIL import Image
 import torch
 from torchvision import transforms
-import re
 
 # MODELES
 INGREDIENT_MODEL_ID = "stchakman/Fridge_Items_Model"
-RECIPE_MODEL_ID     = "flax-community/t5-recipe-generation"
+RECIPE_MODEL_ID = "flax-community/t5-recipe-generation"
 
 # PIPELINES
 ingredient_classifier = pipeline(
     "image-classification",
     model=INGREDIENT_MODEL_ID,
     device=0 if torch.cuda.is_available() else -1,
-    top_k=5
+    top_k=4
 )
+
 recipe_generator = pipeline(
     "text2text-generation",
     model=RECIPE_MODEL_ID,
-    device=0 if torch.cuda.is_available() else -1,
+    device=0 if torch.cuda.is_available() else -1
 )
 
 # AUGMENTATION
@@ -29,59 +29,30 @@ augment = transforms.Compose([
     transforms.ColorJitter(brightness=0.2, contrast=0.2),
 ])
 
-def parse_recipe_text(raw: str):
-    """
-    Parse le texte brut retourne par le modele en JSON { ingredients: [...], steps: [...] }.
-    """
-    raw = raw.replace("ingredients: ingredients:", "ingredients:")
-    # On cherche la section ingredients et directions
-    ing_match = re.search(r"ingredients:\s*(.*?)\s*directions:", raw, re.IGNORECASE | re.DOTALL)
-    dir_match = re.search(r"directions:\s*(.+)$", raw, re.IGNORECASE | re.DOTALL)
-
-    # ingredients
-    ingredients = []
-    if ing_match:
-        ing_str = ing_match.group(1).strip()
-        parts = re.findall(r"(\d+\s+[^\d,\.]+(?: [^\d,\.]+)*)", ing_str)
-        if parts:
-            ingredients = [p.strip() for p in parts]
-        else:
-            ingredients = [i.strip() for i in ing_str.split(",") if i.strip()]
-
-    # etapes
-    steps = []
-    if dir_match:
-        steps_str = dir_match.group(1).strip()
-        for s in re.split(r"\.\s*", steps_str):
-            s = s.strip()
-            if s:
-                steps.append(s.endswith(".") and s or s + ".")
+# FONCTION PRINCIPALE
+def generate_recipe(image: Image.Image):
+    yield "🔄 Traitement de l'image... Veuillez patienter."
 
-    return {"ingredients": ingredients, "steps": steps}
-
-def generate_recipe_json(image: Image.Image):
-    # Aug
+    # Augmentation
     image_aug = augment(image)
+
+    # Classification
     results = ingredient_classifier(image_aug)
     ingredients = [res["label"] for res in results]
     ingredient_str = ", ".join(ingredients)
-    prompt = f"Ingredients: {ingredient_str}. Recipe:"
-    raw = recipe_generator(prompt, max_new_tokens=256, do_sample=True)[0]["generated_text"]
 
-    parsed = parse_recipe_text(raw)
-    return {
-        "detected_ingredients": ingredients,
-        "recipe_text": raw,
-        "ingredients_parsed": parsed["ingredients"],
-        "steps": parsed["steps"]
-    }
+    yield f"🥕 Ingrédients détectés : {ingredient_str}\n\n🍳 Génération de la recette..."
+    prompt = f"Ingredients: {ingredient_str}. Recipe:"
+    recipe = recipe_generator(prompt, max_new_tokens=256, do_sample=True)[0]["generated_text"]
+    yield f"### 🥕 Ingrédients détectés :\n{ingredient_str}\n\n### 🍽️ Recette générée :\n{recipe}"
 
+# INTERFACE
 interface = gr.Interface(
-    fn=generate_recipe_json,
-    inputs=gr.Image(type="pil", label="📷 Image de vos ingredients"),
-    outputs="json",
-    title="🧑‍🍳 Generateur de Recettes (JSON) 🧑‍🍳",
-    description="Depose une image d'ingredients, recupere directement un JSON structure avec ingredients et etapes.",
+    fn=generate_recipe,
+    inputs=gr.Image(type="pil", label="📷 Image de vos ingrédients"),
+    outputs=gr.Markdown(),
+    title="🥕 Générateur de Recettes 🧑‍🍳",
+    description="Dépose une image d'ingrédients pour obtenir une recette automatiquement générée à partir d'un modèle IA.",
     allow_flagging="never"
 )