HOU Yuenan
HOU Yuenan
SCNN has a message passing module which works extremely well in long-continuous object detection like lane detection. However, since message passing module accounts for around 30% of the total feedforward...
GeForce GTX TITAN X.
@BoSeal You can easily modify the codes of SCNN-Tensorflow to adapt to Cityscapes dataset as long as you change the input image size, some parameters (fc layers) and the dataloader.
The ground-truth labels for crossroads contain no lanes. So there is no TP and FN. And FP is the number of detected lanes in this condition.
@yellowYuga，我觉得这里使用relu只是起一个非线性映射的作用，这里的迭代是为了将信息从上向下，从下向上，从左到右，从右到左传递开来，如果是没有用的信息（负数），那就不传递
@passion3394 , please refer to [How to train SCNN on my own dataset #7](https://github.com/XingangPan/SCNN/issues/7)
@passion3394 , the performance depends on the proportion of the number of lanes and the difficulty of road scenarios. I have not tested models predicting more than 4 lanes.
@passion3394 , you can also refer to [this repo](https://github.com/cardwing/Codes-for-Lane-Detection).
@passion3394 , you are welcome.
@ytzhao , @ding-hai-tao , I will implement a tensorflow version of SCNN after 11.16. If you are interested in this, you can refer to [Codes-for-Lane-Detection](https://github.com/cardwing/Codes-for-Lane-Detection) where I will put my...