Michael liu
Michael liu
> Hi @liuheng92 I see that you reopened the issue, have you managed to figure it out? no, I can't handle it
> > 不知怎么读取mdb格式，为啥不弄成json格式呢 > > Using `lmdb` format just to avoid out-of-memory problem when using DDP for training. You can also use `json` format annotation file like icdar2015 and icdar2017....
> > 不知怎么读取mdb格式，为啥不弄成json格式呢 > > Using `lmdb` format just to avoid out-of-memory problem when using DDP for training. You can also use `json` format annotation file like icdar2015 and icdar2017....
> Sorry, MMOCR is not supported yet. We expect to support mmocr in July What's the different between mmocr and other repos, I mean how I can apply mmocr using...
> What type of compression algorithm do you want to use? KD? NAS? Pruning? If it is OneShot NAS or KD, it may be easier. I want to use autoslim,...
> same thing is happening to me. Which version of tensorflow and efficientnet are you using? > > I'm using tensorflow 1.15.0 and efficientnet 1.0.0 I am using tensorflow 1.13,...
> When you say slow training speed do you mean examples per second or epochs to convergence? > > I'm seeing slightly more examples per second than resnet50, but many...
检查一下数据是否按照格式存放的
这个算法我没有转pb，你可以自己打印节点名
> > > 八个数字后面加文本string，因为是训练检测模型，这个string我写的是null（随便写都可以），可以正常训练。 > > > […](#) > > > On Tue, Nov 19, 2019 at 2:11 PM Asuna88 _**@**_.***> wrote: 这个模型我down下来跑过，里面的多进程会导致内存溢出，每次step内存持续变大，初步估计自动回收机制无法实现实时回收，手动gc也失效。我将generator去掉，单进程跑，不会出现溢出，目前还没完美的解决办法 你好，，我是这么处理多线程转单线程的，不知道对否，大神请指点。 在data_util.py文件里面，注释掉下面几句话，然后再补充了三句话，请问大神这样做对否。 （ 请问如果写成 workers=1, max_queue_size =2...