Results 8 comments of CHENHUI

大佬能大概说说代码的思路吗,小弟太菜没看明白这里代码,怎么实现的

> > 大佬能大概说说代码的思路吗,小弟太菜没看明白这里代码,怎么实现的 > > 自注意力机制 Q K 相乘不是出来一个矩阵嘛? 然后比如 i行 j列这个元素,代表第i个token和第j个token之间的关系。然后来自不同窗口的两个token应该没关系,所以应该强行置0。 > ```python > attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2) > ``` > 这句话就是给 算出来的矩阵标序号, 算出来来自一个窗口为0, 不同窗口不为0。 不为0的给原矩阵对应位置-100, 这样softmax出来这里就接近0, 也就达到了前面说的强行置0的效果....

> https://github.com/ayooshkathuria/YOLO_v3_tutorial_from_scratch/blob/8264dfba39a866998b8936a24133f41f12bfbdb7/util.py#L59 > > I have a question since yolov3 has anchors for all three different scales. (as they mentioned in paper). Why again we need to down sample the...

In fact, in my opinion, if you get an exact same result, that's what it shouldn't be.

> 没用,放弃治疗了。还是用原本的clash吧,之前看这个界面好看,还是算了

> Firstly, the data sample strategy is **bagging** by default. > > > So, GOSS algorithm won't work if you don't set core parameter **data_sample_strategy = goss**. > > Secondly,...