Trained LaneATT model

[daeunni]

Hi,
do you have trained LaneATT CUlane & TUSimple models? :) (like resnet18s_lstr_culane_20210722.pt)
where can I find them in this repo?

[voldemortX]

@daeunni LaneATT is currently supported on CULane. You can find the models in MODEL_ZOO.md

e.g., resnet18: https://drive.google.com/file/d/17Ve7RiqxHK4aEXOt7MJzycj6mvvrieZ1/view?usp=sharing

For tusimple, there are some technical reproducibility issues: #98

[voldemortX]

Remember to download anchors from Tabelini's repo:

pytorch-auto-drive/configs/lane_detection/laneatt/resnet18_culane.py

Line 141 in 8cc1d30

# https://github.com/lucastabelini/LaneATT/raw/main/data/culane_anchors_freq.pt

[daeunni]

Thanks a lot! @voldemortX :)

btw, I encountered this error. How can I compile line_nms ?
NameError: name 'line_nms' is not defined

I tried #103 method but I can't install the laneatt nms packages.

[voldemortX]

@daeunni You mean even the install-mode nms in LaneATT repo can't be installed? Perhaps use verbose=True here and see if there is a compiling issue?

[daeunni]

Ohh yes ! After I searched some compile error about install-mode nms, I found there was a Ninja error like below.
RuntimeError: Ninja is required to load C++ extension

So I solved that problem using this. It will be helpful to someone ! :3 (It's also related to #103)

wget https://github.com/ninja-build/ninja/releases/download/v1.8.2/ninja-linux.zip
sudo unzip ninja-linux.zip -d /usr/local/bin/
sudo update-alternatives --install /usr/bin/ninja ninja /usr/local/bin/ninja 1 --force 

[voldemortX]

@daeunni Thanks for the solution! Does pip install ninja work?

[daeunni]

Yep that works to me

[voldemortX]

Well. I think it may be time to add ninja to requirements.txt.