enazoe / yolo-tensorrt

TensorRT8.Support Yolov5n,s,m,l,x .darknet -> tensorrt. Yolov4 Yolov3 use raw darknet *.weights and *.cfg fils. If the wrapper is useful to you,please Star it.
MIT License
1.19k stars 315 forks source link

你好,请问一下,用int8精度的模式怎样使用? #8

Closed 121649982 closed 4 years ago

121649982 commented 4 years ago

需要自己加上些图像吗?

enazoe commented 4 years ago

@121649982 config.inference_precison = INT8; config.calibration_image_list_file_txt = "../configs/calibration_images.txt"; int8校准的时候需要一些图像,用calibration_image_list_file_txt指定图片的路径

121649982 commented 4 years ago

@121649982 config.inference_precison = INT8; config.calibration_image_list_file_txt = "../configs/calibration_images.txt"; int8校准的时候需要一些图像,用calibration_image_list_file_txt指定图片的路径

请问一下,如果用自己训练的模型,这些图像可以是用于训练的吗?感谢

enazoe commented 4 years ago

@121649982 就是需要你训练darknet模型时候的图片作为trt校准的图片,不需要全部,一个子集就行,几百张

121649982 commented 4 years ago

@enazoe 好的,谢谢了

enazoe commented 4 years ago

@121649982 注意这个子集要是训练集的均匀采样,如果觉得还不错点个star。。。。

121649982 commented 4 years ago

好的,感谢,你的代码真的很赞,我在windows上跑起来了

Stars-Chan commented 3 years ago

在Linux上注意把calibration_images.txt文件里面的”\“换成”/“

wangxudong-cq commented 2 years ago

@121649982 config.inference_precison = INT8; config.calibration_image_list_file_txt = "../configs/calibration_images.txt"; int8校准的时候需要一些图像,用calibration_image_list_file_txt指定图片的路径

这个数据集是自动下载的吗?

enazoe commented 2 years ago

@121649982 config.inference_precison = INT8; config.calibration_image_list_file_txt = "../configs/calibration_images.txt"; int8校准的时候需要一些图像,用calibration_image_list_file_txt指定图片的路径

这个数据集是自动下载的吗? https://github.com/enazoe/yolo-tensorrt/issues/8#issuecomment-607185879 https://github.com/enazoe/yolo-tensorrt/releases/tag/calibration_images.txt

wangxudong-cq commented 2 years ago

@enazoe thank you! I have copied the images from the Coco training dataset。