三级实操速查 — 3.1 Excel + 3.2 图片识别

# ===== 3.2.1 ResNet 图片分类 — 完整代码模板 ===== import onnxruntime as ort import numpy as np import scipy.special from PIL import Image # 预处理函数（题目已提供，不用填） def preprocess_image(image, resize_size=256, crop_size=224, mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]): image = image.resize((resize_size, resize_size), Image.BILINEAR) w, h = image.size left = (w - crop_size) / 2 top = (h - crop_size) / 2 image = image.crop((left, top, left + crop_size, top + crop_size)) image = np.array(image).astype(np.float32) image = image / 255.0 image = (image - mean) / std image = np.transpose(image, (2, 0, 1)) image = image.reshape((1,) + image.shape) return image # ① 加载模型 (2分) session = ort.InferenceSession('resnet.onnx') # 加载标签 with open('labels.txt') as f: labels = [line.strip() for line in f.readlines()] input_name = session.get_inputs()[0].name output_name = session.get_outputs()[0].name # ② 加载图片 (2分) image = Image.open('img_test.jpg').convert('RGB') # ③ 预处理图片 (2分) processed_image = preprocess_image(image) processed_image = processed_image.astype(np.float32) # ④ 执行推理 (2分) output = session.run([output_name], {input_name: processed_image})[0] # ⑤ softmax (2分) probabilities = scipy.special.softmax(output, axis=-1) # ⑥ 取 Top5 (3分) top5_idx = np.argsort(probabilities[0])[-5:][::-1] top5_prob = probabilities[0][top5_idx] print("Top 5 predicted classes:") for i in range(5): print(f"{i+1}: {labels[top5_idx[i]]} - Probability: {top5_prob[i]}")

# ===== 3.2.2 MNIST 手写数字 — 完整代码模板 ===== import onnxruntime import numpy as np from PIL import Image # ① 加载模型 (2分) ort_session = onnxruntime.InferenceSession('mnist.onnx') # ② 加载图片 + 转灰度 (2分) image = Image.open("img_test.png").convert('L') # ③ resize (2分) image = image.resize((28, 28)) # ④ 转 numpy (2分) image_array = np.array(image, dtype=np.float32) # ⑤ 添加 batch 维度 (2分) image_array = np.expand_dims(image_array, axis=0) # ⑥ 添加 channel 维度 (2分) image_array = np.expand_dims(image_array, axis=0) # ⑦ 获取输入名 (2分) ort_inputs = {ort_session.get_inputs()[0].name: image_array} # ⑧ 执行推理 (2分) ort_outs = ort_session.run(None, ort_inputs) # ⑨ 取最大概率的类别 (2分) predicted_class = np.argmax(ort_outs) print(f"Predicted class: {predicted_class}")

# ===== 3.2.3 情感识别 — 完整代码模板 ===== import numpy as np from PIL import Image import onnxruntime as ort # 预处理函数（题目已提供） def preprocess(image_path): input_shape = (1, 1, 64, 64) img = Image.open(image_path).convert('L') img = img.resize((64, 64), Image.ANTIALIAS) img_data = np.array(img, dtype=np.float32) img_data = np.expand_dims(img_data, axis=0) img_data = np.expand_dims(img_data, axis=1) assert img_data.shape == input_shape return img_data # ① 情感映射表 (3分) — 8 类情感 emotion_table = {0: 'Neutral', 1: 'Happy', 2: 'Sad', 3: 'Surprise', 4: 'Fear', 5: 'Disgust', 6: 'Anger', 7: 'Contempt'} # ② 加载模型 (3分) ort_session = ort.InferenceSession('emotion-ferplus.onnx') # ③ 预处理图片 (3分) input_data = preprocess('img_test.png') ort_inputs = {ort_session.get_inputs()[0].name: input_data} # ④ 执行推理 (3分) ort_outs = ort_session.run(None, ort_inputs) # ⑤ 取预测类别 (3分) predicted_label = np.argmax(ort_outs[0]) # ⑥ 查情感名 (3分) predicted_emotion = emotion_table[predicted_label] print(f"Predicted emotion: {predicted_emotion}")

# ===== 3.2.4 花卉分类 — 完整代码模板 ===== import onnxruntime as ort import numpy as np import scipy.special from PIL import Image # 预处理函数（同 3.2.1） def preprocess_image(image, resize_size=256, crop_size=224, mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]): image = image.resize((resize_size, resize_size), Image.BILINEAR) w, h = image.size left = (w - crop_size) / 2 top = (h - crop_size) / 2 image = image.crop((left, top, left + crop_size, top + crop_size)) image = np.array(image).astype(np.float32) image = image / 255.0 image = (image - mean) / std image = np.transpose(image, (2, 0, 1)) image = image.reshape((1,) + image.shape) return image # ① 加载模型 (2分) session = ort.InferenceSession('flower-detection.onnx') # ② 加载标签 (2分) with open('labels.txt') as f: labels = [line.strip() for line in f.readlines()] input_name = session.get_inputs()[0].name output_name = session.get_outputs()[0].name # ③ 加载图片 (2分) image = Image.open('flower_test.png').convert('RGB') # ④ 预处理 (2分) processed_image = preprocess_image(image) processed_image = processed_image.astype(np.float32) # ⑤ 推理 (2分) output = session.run([output_name], {input_name: processed_image})[0] # ⑥ softmax (2分) accuracy = scipy.special.softmax(output, axis=-1) # ⑦ 取预测索引 predicted_idx = np.argmax(accuracy) # ⑧ 转百分比 prob_percentage = accuracy[0][predicted_idx] * 100 # ⑨ 取标签名 predicted_label = labels[predicted_idx] print(f"Predicted class: {predicted_label}, Accuracy: {prob_percentage:.2f}%")

# ===== 3.2.5 人脸检测 — 填空关键行 ===== # 已有导入和 predict 函数（不用填） # ① 加载标签 (2分) class_names = [name.strip() for name in open('voc-model-labels.txt').readlines()] # ② 加载模型 (2分) ort_session = ort.InferenceSession('version-RFB-320.onnx') # ③ 获取输入名 (2分) input_name = ort_session.get_inputs()[0].name result_path = "./detect_imgs_results_onnx" threshold = 0.7 path = "imgs" sum = 0 # ④ 创建目录 (2分) if not os.path.exists(result_path): os.makedirs(result_path) listdir = os.listdir(path) for file_path in listdir: img_path = os.path.join(path, file_path) # ⑤ 读取图片 (2分) orig_image = cv2.imread(img_path) image = cv2.cvtColor(orig_image, cv2.COLOR_BGR2RGB) # ⑥ resize (2分) image = cv2.resize(image, (320, 240)) # ⑦ 均值 (2分) image_mean = np.array([127, 127, 127]) image = (image - image_mean) / 128 image = np.transpose(image, [2, 0, 1]) # ⑧ expand_dims (1分) image = np.expand_dims(image, axis=0) image = image.astype(np.float32) time_time = time.time() # ⑨ 推理 (2分) confidences, boxes = ort_session.run(None, {input_name: image}) print("cost time:{}".format(time.time() - time_time)) boxes, labels, probs = predict(orig_image.shape[1], orig_image.shape[0], confidences, boxes, threshold) for i in range(boxes.shape[0]): box = boxes[i, :] label = f"{class_names[labels[i]]}: {probs[i]:.2f}" cv2.rectangle(orig_image, (box[0], box[1]), (box[2], box[3]), (255, 255, 0), 4) cv2.imwrite(os.path.join(result_path, file_path), orig_image) sum += boxes.shape[0] print("sum:{}".format(sum))