Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,55 +1,44 @@
|
|
| 1 |
-
import json
|
| 2 |
-
from tensorflow.keras.models import load_model
|
| 3 |
-
from tensorflow.keras.preprocessing.image import img_to_array
|
| 4 |
-
from PIL import ImageOps, Image
|
| 5 |
-
import numpy as np
|
| 6 |
import gradio as gr
|
|
|
|
|
|
|
| 7 |
|
| 8 |
-
#
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
CLASS_ORDER = json.load(open("label_map.json"))
|
| 14 |
-
|
| 15 |
-
# ——————————————————————
|
| 16 |
-
# 2) PRÉ-PROCESSAMENTO robusto para fotos de telemóvel
|
| 17 |
-
IMG_SIZE = (224, 224)
|
| 18 |
-
MODEL_PATH= "cropvision_model.keras"
|
| 19 |
-
model = load_model(MODEL_PATH)
|
| 20 |
-
import hashlib, pathlib
|
| 21 |
-
# debug: imprime MD5 e número de parâmetros
|
| 22 |
-
md5 = hashlib.md5(pathlib.Path(MODEL_PATH).read_bytes()).hexdigest()
|
| 23 |
-
print("📝 MD5 do modelo carregado:", md5)
|
| 24 |
-
print("🔢 camadas:", len(model.layers), "-- params:", model.count_params())
|
| 25 |
|
|
|
|
|
|
|
| 26 |
|
| 27 |
def predict(image: Image.Image):
|
| 28 |
-
# corrige
|
| 29 |
-
|
| 30 |
-
img = ImageOps.fit(img, IMG_SIZE, Image.Resampling.LANCZOS)
|
| 31 |
-
|
| 32 |
-
# normaliza e infere
|
| 33 |
-
arr = img_to_array(img)/255.0
|
| 34 |
-
arr = np.expand_dims(arr, 0)
|
| 35 |
-
probs = model.predict(arr)[0]
|
| 36 |
-
|
| 37 |
-
# escolhe a classe e devolve também todas as probabilidades
|
| 38 |
-
idx = int(np.argmax(probs))
|
| 39 |
-
label = CLASS_ORDER[idx]
|
| 40 |
-
mapping = {CLASS_ORDER[i]: float(probs[i]) for i in range(len(probs))}
|
| 41 |
-
|
| 42 |
-
return label, f"{mapping}"
|
| 43 |
|
| 44 |
-
#
|
| 45 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 46 |
demo = gr.Interface(
|
| 47 |
fn=predict,
|
| 48 |
-
inputs=gr.Image(type="pil", label="Carrega folha"),
|
| 49 |
-
outputs=[
|
| 50 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 51 |
)
|
| 52 |
|
| 53 |
-
if __name__=="__main__":
|
| 54 |
demo.launch()
|
| 55 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
+
from PIL import Image
|
| 3 |
+
from transformers import pipeline
|
| 4 |
|
| 5 |
+
# 1) Cria o classificador zero‐shot com CLIP
|
| 6 |
+
classifier = pipeline(
|
| 7 |
+
task="zero-shot-image-classification",
|
| 8 |
+
model="openai/clip-vit-base-patch32"
|
| 9 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
|
| 11 |
+
# 2) Define as labels
|
| 12 |
+
LABELS = ["Healthy", "Leaf Blight", "Black Rot", "ESCA"]
|
| 13 |
|
| 14 |
def predict(image: Image.Image):
|
| 15 |
+
# Opcional: corrige EXIF e redimensiona como antes
|
| 16 |
+
image = image.convert("RGB").resize((224,224))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
|
| 18 |
+
# Zero‐shot classification
|
| 19 |
+
res = classifier(image, candidate_labels=LABELS)
|
| 20 |
+
# res é lista de dicts: [{"label":..., "score":...}, ...]
|
| 21 |
+
|
| 22 |
+
# Mapeia para texto ordenado
|
| 23 |
+
probs = {item["label"]: float(item["score"]) for item in res}
|
| 24 |
+
# Escolhe o mais provável
|
| 25 |
+
best = max(probs, key=probs.get)
|
| 26 |
+
|
| 27 |
+
# Formata saída
|
| 28 |
+
prob_lines = "\n".join(f"{lbl}: {probs[lbl]:.2f}" for lbl in LABELS)
|
| 29 |
+
return best, prob_lines
|
| 30 |
+
|
| 31 |
+
# 3) Interface Gradio
|
| 32 |
demo = gr.Interface(
|
| 33 |
fn=predict,
|
| 34 |
+
inputs=gr.Image(type="pil", label="Carrega a folha"),
|
| 35 |
+
outputs=[
|
| 36 |
+
gr.Textbox(label="Classe predita"),
|
| 37 |
+
gr.Textbox(label="Probabilidades (0–1)")
|
| 38 |
+
],
|
| 39 |
+
title="CropVision (Backup CLIP Zero-Shot)",
|
| 40 |
+
description="Usa CLIP zero-shot para classificar folhas em Healthy, Leaf Blight, Black Rot ou ESCA"
|
| 41 |
)
|
| 42 |
|
| 43 |
+
if __name__ == "__main__":
|
| 44 |
demo.launch()
|
|
|