yehao
yehao
You should clone weiliu's ssd-caffe version.
@shudct NVcaffe supports for float16.
@kimys1084 Can you post your demo_webcam.py code?
I will check code @kimys1084, please wait!
In your first graph, detection_eval = 5.5%, this is mAP
It is common when training model, you can try to train for several times and you will find mAp of each training is different, you can choose a better-trained model...
@ArtyZe 请教下你这个L1正则化是针对conv weights还是针对BN的scale值做的？我正在实验一个模型剪枝方法，参考的是这篇文章 > Learning efficient convolution networks through network slimming(ICCV 2017)
> use plus =1, the error occurs: centroids.append(boxes[centroid_index]) TypeError: only integer scalar arrays can be converted to a scalar index just modify `centroids.append(boxes[centroid_index])` to `centroids.append(boxes[int(centroid_index)])`