wujiyang

lfw的结果高并说明不了什么，能在megaface上刷高分数才行，还有就是要看你使用的什么训练集

I suggested that you should use face detection and face alignment algorithms to do data preprocess first, not just resize only.

Well, you can use this tool: [https://github.com/deepinsight/insightface/tree/master/src/align](https://github.com/deepinsight/insightface/tree/master/src/align)

Did you use the input size of 112*112 ? If not, you should modify the size of last convolutional feature map.

25088 = 512 * 7 * 7； 32768 = 512 * 8 * 8； It seems that you are using 120 * 120 input. Can you give more details on...

我使用的是MTCNN检测对齐算法，开源的MTCNN代码网上很多，检测到关键点以后采用相似性变换进行对齐。比如 [kpzhang93/MTCNN_face_detection_alignment](https://github.com/kpzhang93/MTCNN_face_detection_alignment) 以及 [deepinsight/insightface](https://github.com/deepinsight/insightface/tree/master/src/align) 都有源代码提供

All setting detailes are in train.py and all network deatils are in backbone/xx.py, you can review the code to get what you want. (You'd better check the code to make...

@satakiolo 稍后我会放到百度云

@satakiolo 链接：https://pan.baidu.com/s/1eMpfz2jCbZR2XO91hgFnDA 提取码：l2zk

那个可以忽略