ZJ
@lsy17096535 [https://github.com/zeusees/HyperFT/blob/master/Prj-Linux/colotracker.cpp#L219](https://github.com/zeusees/HyperFT/blob/master/Prj-Linux/colotracker.cpp#L219)我在实验中发现这个函数的frame2_features坐标好多是负值，然后最后产生的pcount（好点数）为0，导致真个tracking函数返回的track box是没有的。
编译过程用的是opencv3.4，ncnn 为20190320，gcc 4.8.5。 编译过程shared_std报错，加了头文件 `#include ` ，输入视频流为本地视频
@hackeritchy 在寻找光流特征点的时候你可以试试大一点的搜索尺寸，我这边改了之后是不报错了。可以的话麻烦告知一下我
@zeven11 Have you solved your problem? I have the same error `writer_dict['writer'].add_graph(model, (dump_input, ))`
佩服大佬的复现，请问这个新的预训练模型上的训练有最新结果么，跟论文比怎么样。
后续你们会在snet49上进行imagenet预训练么
赞！大佬voc\coco测试代码什么时候放出啊
Soga. If I change the config.yaml, add the batchsize,data_root ..., multi-task experiments should also be started . Is right?
Ok. I will try it. Another question，I noticed that the input size of the model is 224. Did you try to compare the input performance of 112 dimensions?
> For example: > train: > rand_seed: 0 > batch_size: [59, 5] # ratio ~ ms1m : webface > loss_weight: [0.92, 0.08] # ratio ~ ms1m : webface > base_lr:...