Update app.py
Browse files
app.py
CHANGED
|
@@ -1,74 +1,121 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
import tensorflow as tf
|
| 3 |
import numpy as np
|
|
|
|
| 4 |
from PIL import Image
|
| 5 |
|
| 6 |
# -----------------------------
|
| 7 |
-
# LOAD MODEL
|
| 8 |
# -----------------------------
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
|
|
|
|
|
|
|
|
|
| 12 |
|
| 13 |
# -----------------------------
|
| 14 |
-
#
|
| 15 |
# -----------------------------
|
| 16 |
-
|
| 17 |
-
|
|
|
|
|
|
|
|
|
|
| 18 |
|
| 19 |
-
|
|
|
|
| 20 |
|
| 21 |
-
|
| 22 |
-
#
|
| 23 |
-
|
| 24 |
-
|
| 25 |
|
| 26 |
# -----------------------------
|
| 27 |
-
#
|
| 28 |
# -----------------------------
|
| 29 |
-
def
|
| 30 |
-
# Convert correctly based on channels
|
| 31 |
if C == 1:
|
| 32 |
-
image = image.convert("L")
|
| 33 |
else:
|
| 34 |
image = image.convert("RGB")
|
| 35 |
|
| 36 |
-
# Resize EXACTLY to training size
|
| 37 |
image = image.resize((W, H))
|
|
|
|
| 38 |
|
| 39 |
-
img_array = np.array(image, dtype=np.float32)
|
| 40 |
-
|
| 41 |
-
# If grayscale, add channel dim
|
| 42 |
if C == 1:
|
| 43 |
-
|
| 44 |
|
| 45 |
-
|
| 46 |
-
img_array = np.expand_dims(img_array, axis=0)
|
| 47 |
|
| 48 |
-
preds =
|
| 49 |
idx = int(np.argmax(preds))
|
| 50 |
|
| 51 |
return {
|
| 52 |
-
"
|
|
|
|
| 53 |
"confidence": float(np.max(preds)),
|
| 54 |
"probabilities": {
|
| 55 |
-
|
| 56 |
-
for i in range(len(
|
| 57 |
-
}
|
| 58 |
-
"model_input_shape": f"{H}x{W}x{C}"
|
| 59 |
}
|
| 60 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 61 |
# -----------------------------
|
| 62 |
# GRADIO APP
|
| 63 |
# -----------------------------
|
| 64 |
demo = gr.Interface(
|
| 65 |
fn=predict,
|
| 66 |
-
inputs=
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 67 |
outputs=gr.JSON(label="Prediction Output"),
|
| 68 |
-
title="O-Ring Defect Detection",
|
| 69 |
-
description="CNN
|
| 70 |
)
|
| 71 |
|
| 72 |
-
demo.launch(
|
| 73 |
-
ssr_mode=False
|
| 74 |
-
)
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
import tensorflow as tf
|
| 3 |
import numpy as np
|
| 4 |
+
import onnxruntime as ort
|
| 5 |
from PIL import Image
|
| 6 |
|
| 7 |
# -----------------------------
|
| 8 |
+
# LOAD CNN MODEL (GOOD / BREAKAGE)
|
| 9 |
# -----------------------------
|
| 10 |
+
CNN_MODEL_PATH = "o_ring_cnn_classifier (1).keras"
|
| 11 |
+
cnn_model = tf.keras.models.load_model(CNN_MODEL_PATH)
|
| 12 |
+
_, H, W, C = cnn_model.input_shape
|
| 13 |
+
|
| 14 |
+
CNN_CLASSES = ["good", "breakage"]
|
| 15 |
+
print("✅ CNN model loaded")
|
| 16 |
|
| 17 |
# -----------------------------
|
| 18 |
+
# LOAD ONNX MODEL (PAREGI / PELISE)
|
| 19 |
# -----------------------------
|
| 20 |
+
ONNX_MODEL_PATH = "best.onnx"
|
| 21 |
+
onnx_session = ort.InferenceSession(
|
| 22 |
+
ONNX_MODEL_PATH,
|
| 23 |
+
providers=["CPUExecutionProvider"]
|
| 24 |
+
)
|
| 25 |
|
| 26 |
+
onnx_input_name = onnx_session.get_inputs()[0].name
|
| 27 |
+
onnx_output_name = onnx_session.get_outputs()[0].name
|
| 28 |
|
| 29 |
+
ONNX_CLASSES = ["paregi", "pelise"]
|
| 30 |
+
ONNX_IMG_SIZE = 640 # ⚠️ change ONLY if your model was trained differently
|
| 31 |
+
|
| 32 |
+
print("✅ ONNX model loaded")
|
| 33 |
|
| 34 |
# -----------------------------
|
| 35 |
+
# CNN PREDICTION
|
| 36 |
# -----------------------------
|
| 37 |
+
def predict_cnn(image: Image.Image):
|
|
|
|
| 38 |
if C == 1:
|
| 39 |
+
image = image.convert("L")
|
| 40 |
else:
|
| 41 |
image = image.convert("RGB")
|
| 42 |
|
|
|
|
| 43 |
image = image.resize((W, H))
|
| 44 |
+
img = np.array(image, dtype=np.float32) / 255.0
|
| 45 |
|
|
|
|
|
|
|
|
|
|
| 46 |
if C == 1:
|
| 47 |
+
img = np.expand_dims(img, axis=-1)
|
| 48 |
|
| 49 |
+
img = np.expand_dims(img, axis=0)
|
|
|
|
| 50 |
|
| 51 |
+
preds = cnn_model.predict(img)[0]
|
| 52 |
idx = int(np.argmax(preds))
|
| 53 |
|
| 54 |
return {
|
| 55 |
+
"model": "CNN Classifier",
|
| 56 |
+
"prediction": CNN_CLASSES[idx],
|
| 57 |
"confidence": float(np.max(preds)),
|
| 58 |
"probabilities": {
|
| 59 |
+
CNN_CLASSES[i]: float(preds[i])
|
| 60 |
+
for i in range(len(CNN_CLASSES))
|
| 61 |
+
}
|
|
|
|
| 62 |
}
|
| 63 |
|
| 64 |
+
# -----------------------------
|
| 65 |
+
# ONNX PREDICTION (PAREGI / PELISE)
|
| 66 |
+
# -----------------------------
|
| 67 |
+
def predict_onnx(image: Image.Image):
|
| 68 |
+
image = image.convert("RGB")
|
| 69 |
+
image = image.resize((ONNX_IMG_SIZE, ONNX_IMG_SIZE))
|
| 70 |
+
|
| 71 |
+
img = np.array(image, dtype=np.float32) / 255.0
|
| 72 |
+
|
| 73 |
+
# HWC → CHW
|
| 74 |
+
img = np.transpose(img, (2, 0, 1))
|
| 75 |
+
img = np.expand_dims(img, axis=0)
|
| 76 |
+
|
| 77 |
+
preds = onnx_session.run(
|
| 78 |
+
[onnx_output_name],
|
| 79 |
+
{onnx_input_name: img}
|
| 80 |
+
)[0][0]
|
| 81 |
+
|
| 82 |
+
idx = int(np.argmax(preds))
|
| 83 |
+
|
| 84 |
+
return {
|
| 85 |
+
"model": "ONNX Paregi/Pelise",
|
| 86 |
+
"prediction": ONNX_CLASSES[idx],
|
| 87 |
+
"confidence": float(np.max(preds)),
|
| 88 |
+
"probabilities": {
|
| 89 |
+
ONNX_CLASSES[i]: float(preds[i])
|
| 90 |
+
for i in range(len(ONNX_CLASSES))
|
| 91 |
+
}
|
| 92 |
+
}
|
| 93 |
+
|
| 94 |
+
# -----------------------------
|
| 95 |
+
# MASTER PREDICT FUNCTION
|
| 96 |
+
# -----------------------------
|
| 97 |
+
def predict(image, model_type):
|
| 98 |
+
if model_type == "CNN (Good / Breakage)":
|
| 99 |
+
return predict_cnn(image)
|
| 100 |
+
else:
|
| 101 |
+
return predict_onnx(image)
|
| 102 |
+
|
| 103 |
# -----------------------------
|
| 104 |
# GRADIO APP
|
| 105 |
# -----------------------------
|
| 106 |
demo = gr.Interface(
|
| 107 |
fn=predict,
|
| 108 |
+
inputs=[
|
| 109 |
+
gr.Image(type="pil", label="Upload O-Ring Image"),
|
| 110 |
+
gr.Radio(
|
| 111 |
+
["CNN (Good / Breakage)", "ONNX (Paregi / Pelise)"],
|
| 112 |
+
label="Select Model",
|
| 113 |
+
value="CNN (Good / Breakage)"
|
| 114 |
+
)
|
| 115 |
+
],
|
| 116 |
outputs=gr.JSON(label="Prediction Output"),
|
| 117 |
+
title="O-Ring Defect Detection System",
|
| 118 |
+
description="Supports CNN classification and ONNX Paregi/Pelise defect detection"
|
| 119 |
)
|
| 120 |
|
| 121 |
+
demo.launch(ssr_mode=False)
|
|
|
|
|
|