hpc203 / Ultra-Fast-Lane-Detection-v2-opencv-onnxrun

分别使用OpenCV、ONNXRuntime部署Ultra-Fast-Lane-Detection-v2车道线检测,包含C++和Python两种版本的程序
43 stars 7 forks source link

无法使用自己转换的ONNX推理 #3

Open CrazyMustard-404 opened 1 year ago

CrazyMustard-404 commented 1 year ago

大佬你好,感谢您的开源工作!我在使用你提供ufldv2_culane_res18_320x800.onnx文件在main.py推理时成功,但在使用我自己转换的culane_res18.onnx文件无法成功,出现如下报错:(culane_res18.onnx是利用官方权重转换而来) /home/pc/anaconda3/envs/UFLD/bin/python3.7 /home/pc/FY/Ultra-Fast-Lane-Detection-v2-master/ONNX/opencv/main.py Traceback (most recent call last): File "/home/pc/FY/Ultra-Fast-Lane-Detection-v2-master/ONNX/opencv/main.py", line 99, in net = Ultra_Fast_Lane_Detection_v2(args.modelpath) File "/home/pc/FY/Ultra-Fast-Lane-Detection-v2-master/ONNX/opencv/main.py", line 10, in init self.input_height = int(input_shape[0]) ValueError: invalid literal for int() with base 10: 'res18'