digits training problem


When trying to train a caffe model to detect dog using coco dataset. The mAP value and precision value and recall value are 0. I followed the documentation in ""https://github.com/dusty-nv/jetson-inference#locating-object-coordinates-using-detectnet" to train the data.

You can check the attachment for details.
In the attachment there could only contain part of the training data.
If you want the whole data you can check “https://share.weiyun.com/7fa4f0a935481704c6efbda62508cc87” for details. Which contains training log. and the data set I had used. and the prototxt it generated.

caffe_output.log (867 KB)
dog.tar.gz (1.93 MB)

This is a ML/DL question rather than issue with NGC services. Likely better to post your question on SO or open issue on DIGITS github page. There is also a DIGITS user forum here:



Actually I had tried these both site.

But my problem is still there, so I was wondering if this site could give me some advice on this issue.