Yue Gao

Since we use 16 GPUs (V100 32G) for batch size = 128 per GPU, which means the total batch size is 2048, it is 4 times bigger than your setting....

Since we have provided the tools for convert models to detectron2 and mmdetection formats, you can use the converted models and modify the lines related pretrained model path or resume...

As we noticed from the longer training experiments, there might be saturation for CNN-based model. The gain for longer training is not significant, so we did not report the results...

Hello, since we use docker image to setup the environment, we do not have the requirement.txt file. But you can refer to the Dockerfile to install packages. Mainly you have...

Hello @linusericsson and @xiaoxiong007 The optimization level does affect the performance, for all results of our models are trained with O1 level. There might be quite small differences among different...

I think the reason is that there is no mini-batch training. Could anyone try to implement Mini-Batch as the author's [easy-faster-rcnn](https://github.com/potterhsu/easy-faster-rcnn.pytorch)?

同样遇到这个问题

我手里有 iPhone X 和 Mac 或许我们一起解决这个问题，但我不是很懂 iOS 开发。

@Secbone 网上找的一个美国假日日历 https://p24-calendars.icloud.com/holiday/US_en.ics 在 iOS 上打开链接即可订阅，也能正常显示。

@Secbone 有任何进展么？