Description
问题描述 Please describe your issue
完全无法识别，训练命令：python tools/train.py -c configs/ppyoloe/ppyoloe_plus_crn_s_80e_coco.yml --eval
导出命令：python tools/export_model.py -c configs/ppyoloe/ppyoloe_plus_crn_s_80e_coco.yml -o weights=output/best_model/model.pdparams。使用config = paddle_infer.Config(model_file, params_file)这种模式没有问题。代码如下：time1 = time.time()
模型路径
model_dir = "output_inference/ppyoloe_plus_crn_s_80e_coco"
model_file = os.path.join(model_dir, "model.pdmodel")
params_file = os.path.join(model_dir, "model.pdiparams")
初始化推理配置
config = paddle_infer.Config(model_file, params_file)
config.enable_use_gpu(200, 0)
config.use_gpu()
config.rt
使用 GPU，设置初始显存大小为 100 MB，设备 ID 为 0
如果使用 CPU，可以启用以下代码：
config.disable_gpu()
创建预测器
predictor = paddle_infer.create_predictor(config)
图片预处理
def preprocess(image_path, target_size=(640, 640)):
# 读取图片并转换为 RGB 格式
image = Image.open(image_path).convert("RGB")
# 调整图片尺寸为目标尺寸
image = image.resize(target_size, Resampling.BILINEAR)
# 归一化
image_np = np.array(image).astype("float32") / 255.0
mean = np.array([0, 0, 0], dtype="float32").reshape((1, 1, 3))
std = np.array([1, 1, 1], dtype="float32").reshape((1, 1, 3))
image_np = (image_np - mean) / std
# 转换为 CHW 格式
image_np = image_np.transpose((2, 0, 1)) # HWC -> CHW
image_np = np.expand_dims(image_np, axis=0) # 增加 batch 维度
return image_np
解析输出结果
def parse_output(output_data, threshold=0.75):
results = []
for row in output_data:
batch_id, score, x_min, y_min, x_max, y_max = row
if score > threshold: # 过滤低置信度的结果
# 宽度处理
x_min = float(x_min) * 1280 / 640
x_max = float(x_max) * 1280 / 640
width = x_max - x_min
if width > 330:
continue
results.append({
"box": [x_min, float(y_min) * 720 / 640, x_max, float(y_max) * 720 / 640], # 检测框坐标
"score": float(score), # 置信度分数
"label": int(batch_id) # 假设 batch_id 可以作为类别标签
})
return results
folder_path = r'D:\PythonProject\img_save'
newDir = r'D:\PythonProject\paddle_image'
获取所有文件列表
files = os.listdir(folder_path)
筛选图片文件
image_files = [f for f in files if f.lower().endswith(('.jpg', '.png', '.jpeg', '.bmp'))]
label_data = {
"label": 'null',
"x": -1,
"y": -1,
"score":0.0,
}
labels = ('mfzh', 'dgp', 'mxd', 'qmly', 'kbfl', 'slfy', 'lhbl', 'qnngg', 'hyss', 'xljm', 'klbj', 'jsym')
获取输入和输出句柄
input_names = predictor.get_input_names()
input_tensor1 = predictor.get_input_handle(input_names[0])
scale_factor=np.array([[1.0,1.0]],dtype=np.float32)
input_tensor2 = predictor.get_input_handle(input_names[1])
input_tensor2.copy_from_cpu(scale_factor)
output_names = predictor.get_output_names()
for img_file in image_files:
img_path = os.path.join(folder_path, img_file)
input_data = preprocess(img_path)
# 设置输入数据
input_tensor1.copy_from_cpu(input_data)
# print(output_names)
# 执行推理
predictor.run()
# 获取输出数据
output_tensor = predictor.get_output_handle(output_names[0])
output_data = output_tensor.copy_to_cpu()
#解析并打印结果
results = parse_output(output_data)
if len(results) == 0:
continue
label_data_list = list()
for result in results:
#print(f"Class: {result['label']}, Score: {result['score']:.4f}, Box: {result['box']}")
label_data_d = label_data.copy()
label_data_d["x"] = int(result['box'][0])
label_data_d["y"] = int(result['box'][1])
label_data_d["label"] = labels[result['label']]
label_data_d["score"] = result["score"]
label_data_list.append(label_data_d)
try:
dst = newDir + "/" + img_file
shutil.copy2(img_path, dst)
except Exception as e:
print(f"复制失败: {img_path}, 错误: {e}")
print(img_file)
print(label_data_list)
time2 = time.time()
print(time2 - time1)。