Jie Lei 雷杰

Not sure whether you guys have seen the Tensorflow implementation, it gets even better results than the original paper, see https://github.com/NLPLearn/QANet#results. Hope it helps!

I tested @hackiey 's code, it can reach em 67.515 after 18 epochs, em 68.09 after 56 epochs, w/o EMA.

Sorry for the incompleteness, above are the result of `$ make all`, below are the result of `make test`. CXX src/caffe/test/test_data_layer.cpp src/caffe/test/test_data_layer.cpp: In instantiation of ‘void caffe::DataLayerTest::TestReadCrop() [with TypeParam =...

Thanks for your kindly reply, it's very helpful.

``` lua local dirs = dir.getdirectories(path); table.sort(dirs) ``` It seems that simply add a `table.sort(dirs)` will give the desirable result.

> i found a workaround in my case which might be useful for others. switch from chrome/chromium to mozilla firefox Add my data point: on a MacBook Pro 2018 13',...

Hi @yujiangpu20, Thanks for your interest in our work. I set these ranges based on which specific lengths I care about, it is very objective.

We sample a video frame (an image) every 2 seconds and extract embedding for it.

Is it failed?

Hi @zsxzs, Unfortunately I lost the original feature files as well. You may need to download the original videos and extract them on your own. -- Jie