Linpeng Peng

Results 21 comments of Linpeng Peng

@Aurora11111 ok,按你最新的代码可以跑通,谢谢!

@Aurora11111 您好,请教您最后有测过其他图片吗?我训练的时候模型显示准确率到了90%多,用测试集的图片去做测试,发现准确率也还OK,我数据集是train(20万)和val(5wan).但是用现实场景中的图片做测试,**测试了好几张,完全是错的**,请问你有遇到过这个问题嘛?感觉这两张图片也很相似啊,左边是数据集(都是电子字体)中的图片,右边是真实图片截取的一部分,完全检测不出来,放一大张图片就更检测不出来了 检测结果: ![image](https://user-images.githubusercontent.com/32757213/45751183-30f59900-bc44-11e8-81e2-23fe053a6cd9.png) ![image](https://user-images.githubusercontent.com/32757213/45751497-14a62c00-bc45-11e8-8bd7-51a8c3ce1467.png)

> > > @kargarisaac > > > I am following the file"tf_sampling_compile.sh" to solve it. But I face a porblem :In file included from tensorflow/core/framework/op.h:23:0, > > > from tf_sampling.cpp:6:tensorflow/core/framework/op_def_builder.h:24:10:...

@shengrongjin @leejiajun , I meet the same problem! Could you mind sharing the method of solving this problem? Thanks a lot!

> I am interested in visualizing the point masks for an entire scene, but I am a little confused by the format of the .mat file. Do you have any...

@wenlihaoyu 您好,想问一下fine-tune是将你生成的模型作为预训练模型,然后再训练我们自己的数据集吗?另外如果我想制作自己的label,要怎样进行啊?还有你说的crnn模型输出的概率矩阵环节,应该先过滤后,再进行后续处理,不是很理解这句话啊

非常感谢,难得作者这么耐心!

@beimingmaster 部分手写体可以识别,我测试过了,不知道你的应用场景是怎样的,可以跑作者的模型试一下

> Hi, thanks for your interest in our work. We are working hard to clean the codes for 3D object detection and please stay tuned for the updates. Ok, I...