kpaonaut / VPGNet_for_lane

Vanishing Point Guided Network for lane detection, with post processing
41 stars 21 forks source link

post_process #10

Closed zhangyuelong closed 5 years ago

zhangyuelong commented 5 years ago

1) I recompile your all files in VPGNet/VPGNet_for_lane-master/caltech-lanes-dataset/caltech-lane-detection/src using swig.sh, then i make makefile and generate LaneDetector64 , but idon't know how to use LaneDetector64?could you help me? or i only run ./LaneDetector64 but then opencv error ,but my opencv is not error in fact. 2)I run test case python lane_extension_polyline_for_VPG.py list/input.png -a -o -d output_log then generate well results. 3) I wish we can communicate in Chinese.thanks

kpaonaut commented 5 years ago

You don't have to do anything with LaneDetector64. It is just some middle file generated by swig for py-c++ interface. Just call python lane_extension_polyline_for_MultiNet.py after sourcing swig.sh and you should be fine.

你不需要做makefile等一系列事情。LaneDetector64只是swig产生的中间文件。按照我在首页README写的顺序来做就可以了。首先source swig.sh,然后运行python lane_extension_polyline_for_MultiNet.py就可以。具体这个文件做了哪些事情,代码都在python文件里,建议去看一下源代码。

zhangyuelong commented 5 years ago

谢谢了

zhangyuelong commented 5 years ago

您好,还有两个疑惑: 1.向您确认一下VPGNet文中所述的道路标识分类所使用的数据库应该是文献34的吧?http://www.ananth.in/RoadMarkingDetection.html 2.我想将车道线与道路标识分类一起训练,车道线数据库和道路标识数据库怎么组合比较好?  直接将标签数据放到一个txt中然后生成lmdb格式文件可以吗?

kpaonaut commented 5 years ago

抱歉之前没看到消息。

1.我其实不清楚,我自己训的时候用的是caltech数据库。 2.就这个repo里面的代码而言,不同的任务应该不能一起训练,除非你的数据库里每张图片都同时标注了车道线和路牌的标签。另外,即使这样,如果要混在一起训练,我个人认为会存在数据bias,需要进行额外的数据加强操作。但是具体训练细节就beyond the scope of this repository了,建议单独去做一些research,在这里就不仔细讨论了。