demo.py predict NO BBox on the 2 image in this repo

[Yeeesir]

python3 ./demo.py ./cfg/yolov4-custom.cfg ./weight/yolov4.pth ./data/dog.jpg
这是我用您给出的weight和cfg文件进行inference的命令，您看一下有什么问题吗？

顺便问一句，cfg/里面不同的配置文件的区别是什么，和weight文件有对应关系吗？需要配套使用吗？

[love2learn-7179]

@Dylanbron I think we need to use yolov4.weights instead of yolov4.pth for weightfile. For ./data/dog.jpg, I got below.

data/dog.jpg: Predicted in 0.207703 seconds.
truck: 0.925689
dog: 0.999999
bicycle: 1.000000
save plot results to predictions.jpg

[Yeeesir]

Thank you! I find the weight file on the google drive

[Yeeesir]

Have you trained the model by yourself? I'm wondering how to get .weights file from checkpoints.pth.

[love2learn-7179]

yes, I fine-tuned the model little bit to get updated .pth, then I used the updated .pth in models.py for inference.

[Yeeesir]

Ok, I find that the model used in demo and evaluate_on_coco is darknet, and the model in train and model is Yolov4. Pretrained and checkpoints (pth) file should be used in Yolov4 model. I hope that this info can be updated in README, there may be some people who have not noticed like @Tianxiaomo

[yxwsfz]

@Dylanbron I think we need to use yolov4.weights instead of yolov4.pth for weightfile. For ./data/dog.jpg, I got below.

data/dog.jpg: Predicted in 0.207703 seconds.
truck: 0.925689
dog: 0.999999
bicycle: 1.000000
save plot results to predictions.jpg

您好，我下载了作者提供的参数文件，想运行demo.py 设置了和你说的相同的参数，但是没有结果，我打印了一下中间过程，在model（image）的输出结果就是nan，所以没有输出结果，想问下是什么原因，除了demo.py还有什么其他的地方需要修改吗？