我保存的人工智能模型的终点有问题

Question

我为保存的 AI 模型创建了一个端点 我的模特检查了眼睛的照片来告诉我眼睛是否患有任何疾病 例如 [“白内障”、“糖尿病”、“青光眼”、“正常”] 我的脚本是

import requests
from PIL import Image
import numpy as np
import tensorflow.lite as tflite

model_path = 'efficientnetb3-EyeDisease-96.22.tflite'
interpreter = tflite.Interpreter(model_path)
interpreter.allocate_tensors()

image_path='1145_right.jpeg'
def preprocess_image(image_path):
    # Load and preprocess the image
    image = Image.open(image_path).convert('RGB')
    image = image.resize((224, 224))
    image_array = np.array(image) / 255.0
    image_array = np.expand_dims(image_array, axis=0)
    return image_array.astype(np.float32)

def perform_inference(image_array):
    # Perform inference
    input_tensor = interpreter.tensor(interpreter.get_input_details()[0]['index'])
    input_tensor()[0] = image_array
    interpreter.invoke()
    output_tensor = interpreter.tensor(interpreter.get_output_details()[0]['index'])
    predictions = output_tensor()
    return predictions.tolist()

if __name__ == '__main__':
    data = ['cataract', 'diabetic', 'glaucoma', 'normal'] 
    image_array = preprocess_image(image_path)
    predictions = perform_inference(image_array)
    result = predictions
    index = np.argmax(result)
    print('Inference Result:', data[index])

每次测试模型的结果都返回白内障

ai模型> python x.py 2024-01-18 14:35:50.276823：我tensorflow/core/util/port.cc:113] oneDNN 自定义操作已开启。由于不同计算顺序的浮点舍入误差，您可能会看到略有不同的数值结果。要关闭它们，请设置环境变量

TF_ENABLE_ONEDNN_OPTS=0

。 2024-01-18 14:35:52.465150：我tensorflow/core/util/port.cc:113] oneDNN 自定义操作已开启。由于不同计算顺序的浮点舍入误差，您可能会看到略有不同的数值结果。要关闭它们，请设置环境变量

TF_ENABLE_ONEDNN_OPTS=0

。 信息：为 CPU 创建了 TensorFlow Lite XNNPACK 委托。 警告：尝试使用仅支持静态大小张量的委托和具有动态大小张量的图（张量#299 是动态大小张量）。推理结果：cataract

模型工作正常，但结果错误，我需要帮助才能正确回答我的照片

Answer 1

固定代码是

import torch
import requests
from torchvision import transforms
from PIL import Image
import io

model_path = "public\Aimodel\model_scripted.pt"
model = torch.jit.load(model_path)
image_url = 'https://pbs.twimg.com/media/GEEQErLWAAA4GYJ?format=jpg&name=small'

model.eval()

# Download the image from the URL
response = requests.get(image_url)
image_data = response.content

# Open the image using PIL
image = Image.open(io.BytesIO(image_data))
t = transforms.Compose(
    [
        transforms.Resize((256, 256)),
        transforms.ToTensor(),
    ]
)
image = t(image).unsqueeze(0)

pred_idx = model(image).argmax().item()

classes = ["Cataract", "Diabetic", "Glaucoma", "Normal"]
prediction = classes[pred_idx]
print(f"Prediction: {prediction}")