ylqi007

我把COCO的 train set 转换成 VOC 的格式了，然后可以训练，但是训练完后，COCO没有 2017 的 test set info，怎么测试？ > The test split don't have any annotations (only images). @xxllp @mumu0419 @Caoxiaocao @zhangqinghao0811 你们有在COCO上训练吗，知道如何在COCO上测试吗？还请指教

@KoapT 根据 `outputSize = (W - F + 2P)/S + 1` 这个公式，我觉得添加 padding 只不过是为了让 outputSize 是 inputSize 的 1/2，从而实现 downsample，这种方式进行 conv 的时候，stride = 2。 如果先进行 SAME padding 的卷积，再 2x2 的...

@zhishao @iodncookie @jhfollowgd @CNUyue @Shank2358 @jersonal 你好，想问下你们的 COCO 数据集是怎么处理然后用于训练的？是把 COCO 转换成 PASCAL VOC 的格式吗？ 谢谢

@changfanfan @YunYang1994 请问一下，如果想要在一个新的数据库上训练的话，是应该先用 k-means 进行归一化，生成归一化的 anchor boxes，然后再分别乘以8，16，32，然后放大后的 anchor boxes 文件进行训练吗？

Try this [py-utils](https://github.com/ylqi007/py-utils)

@petit-ami I saw that you were training and testing on PASCAL VOC 2012. I am training and evaluation on PASCAL VOC 2012 right now. 1. I trained on the trainval...

@Fighting-JJ For the first question, I don't think they work in the same way. Because `matching_threshold` should be a constant value of 0.5, while `feat_scores=tf.zeros(shape, dtype=dtype)` (https://github.com/balancap/SSD-Tensorflow/blob/master/nets/ssd_common.py#L57), the initial value...

@ChenhongyiYang I don't know and not sure about that. Sincerely

@stpraha I can understand that the feat_scores new less than 0. What about the accuracy? Does accuracy improve? Can you get the same result with the author's experiment like the...

@stpraha Yes. In my opinion, both the Jaccard score and feat_scores will never be a negative value. Do you have SSD code which can work? Can you share your code?...