fei peng

建议更新一下demo相关的工具，比如bazel有点老的

CPU推理，开启mkldnn报错 RuntimeError: could not execute a primitive

1

执行环境：ubuntu 20.04 + cpu + mkldnn 错误信息： File "/usr/src/app.py", line 160, in invoke ocr_result = ocr_inference(disk_im_fullpath) File "/usr/src/app.py", line 87, in ocr_inference result = ocr.ocr(im, cls=False) File "/usr/local/lib/python3.8/dist-packages/paddleocr/paddleocr.py", line 524,...

inference every frame ,then combine them into a video is a common methond, but this will cause incontinuity。 please, any method get a better result in video?

bug in function match

3

# Problem In [match](https://github.com/amdegroot/ssd.pytorch/blob/8dd38657a3b1df98df26cf18be9671647905c2a0/layers/box_utils.py#L71) function, **which try to ensure every groundtruth boxe can be assigned to a prior boxe**(following code). But there is a problem , if there are same...

With command in readme.md, I got the reulst  `00.jpg`  what is meaning of each col of image ?

Did you make the your model public? I noticed that you have released baseline models, but have you released your own models

Hi, Have you trained the YOLO by your code? what's the performance?

I have three questions about the parameter 'stability 1. What is the range of values for the 'stability' parameter? 2. Does this parameter apply to all models provided by you?...

关于refinedet-resnet

3

您好，我在实现resnet的refinedet（pytorch）。但是几次迭代后，**ODM模块的loss出现了nan**。我在vgg-refinedet上测试没有这种情况，且性能良好，因此想知道是不是基于resnet的refinedet有什么特殊的地方？（比如vgg中的conv4和conv5中的归一化操作） ``` iter 10/120000 || ARM_L Loss: 3.0050 ARM_C Loss: 13.3785 ODM_L Loss: 3.3522 ODM_C Loss: 4.3170 || iter 20/120000 || ARM_L Loss: 4.4867 ARM_C Loss: 14.0901 ODM_L Loss: 4.0826...