flazerain

同样的问题，番剧视频下载不了了

确实GIbox内部的点未必都是师生差距大的点，但是我观察了一下计算结果，这10个点附近还是有很多差异大的点被抑制掉了，不如不用nms。要把bbox预测的影响考虑进去的话，差异大的点的评价也可以综合考虑score 和bbox。 另外实现里面两个batch一起做nms好像也不太对...

> 你是训练前期观察到的还是在训练最后几个epoch观察到的 前期吧，虽然加载了预训练模型

> 两个batch内部做效果更好，我怀疑是如果使用对单张图做loss weight得再调。还有，单张图做就得加for循环搞nms，或者呈平方增长加大IoU矩阵大小 单张图做就得加for循环搞nms，或者呈平方增长加大IoU矩阵大小 这个是什么意思呢？

> can you use caffe model directly instead of use tensorRT model ? > if caffe model is not occur error , you should set shape in .prototxt the same...

> @flazerain > Hi you can convert arcface model to onnx right ?, I convert .,but prelu layer out is error yes,but it is different from the model zoo 's...

but in the view of my comprehension, it seems have some problem if every action has a direct reward. For a random policy at the beginning,the agent can learn to...

but if it randomly chooses a correct action for classification at the first time,it will get 1 reward,because every action have a direct reward.it seems like a supervised signal for...

填了api key 能用gpt-4吗？