issues about test on laneatt

[Auroradsy]

I would like to test my tusimple dataset on laneatt, but I cannot find pretrained laneatt '.pt' model in the modelzoo of this repository(just have two for culane dataset based on resnet18 and resnet34 backbone), so I try to train a new one, but another file needed, 'tusimple_anchors_freq.pt'.

So could you please tell me how to do my test then? Or do you have pretrained laneatt model or suitable 'tusimple_anchors_freq.pt' that can be offered?

Thank you for your such powerful repository and thanks for your patience!

[voldemortX]

The re-implementation of LaneATT has not yet succeeded on TuSimple.

anchors should be available from the LaneATT official repo.

[Auroradsy]

Thank you, did you mean that the implementation for LanaATT in this repository is not completed, even if I got anchors from the official repo?

[voldemortX]

Yes they are untested. As I remember, our accuracy seems to match LaneATT but FP and FN seems off.
Maybe @cedricgsh can provide some further details about LaneATT.

[Auroradsy]

OK! I will have a try after I got anchors!
Thanks again for your patience! :)

[cedricgsh]

The laneatt config has a laneatt_tusimple config file. You can trainging laneatt on tusimple by that config file. But you have to put the tusimple_anchors_freq.pt in main director. The tusimple_anchors_freq.pt can be found in https://github.com/lucastabelini/LaneATT/tree/main/data.
The re-implemented accuracy has almost aligned, but FP and FN are totally different with official result. And I didn' t figure it
out. @Auroradsy

[Auroradsy]

The laneatt config has a laneatt_tusimple config file. You can trainging laneatt on tusimple by that config file. But you have to put the tusimple_anchors_freq.pt in main director. The tusimple_anchors_freq.pt can be found in https://github.com/lucastabelini/LaneATT/tree/main/data. The re-implemented accuracy has almost aligned, but FP and FN are totally different with official result. And I didn' t figure it out. @Auroradsy

Thank you, I got anchors and my training is in progress now!