search

cfzd / Ultra-Fast-Lane-Detection-v2

Ultra Fast Deep Lane Detection With Hybrid Anchor Driven Ordinal Classification (TPAMI 2022)
MIT License
606 stars 99 forks source link

not find tusimple_anno_cache.json #3

Open liujiachang opened 2 years ago

liujiachang commented 2 years ago

I don't find this fie in my data, can you tell me where I can find the tusimple_anno_cache.json, thanks.

cfzd commented 2 years ago

Please follow the instructions here: https://github.com/cfzd/Ultra-Fast-Lane-Detection-v2/blob/master/INSTALL.md#41-tusimple-dataset

liujiachang commented 2 years ago

thanks, because I can run this data in ufldv1, so I don't deal with data for ufldv2.

cfzd commented 2 years ago

@liujiachang In this part, we use a new DALI lib to accelerate the training, so new data preparation is needed.

liujiachang commented 2 years ago

I'm trying to migrate to my own data (intersection surveillance video). Maybe the effect is not ideal due to different angles. Do you have any good suggestions? I can make my own data set, but because the camera is fixed and can't capture multi-angle videos well, I have no idea to solve this problem. This is a example. image

cfzd commented 2 years ago

@liujiachang If I understand you correctly, your problem is that you only have fixed-angle images for training while you need to test on the various angle images? I think you can do some homography transformation as data augmentation, which can simulate the angle change during training.

liujiachang commented 2 years ago

I tried to enhance the data, and it did work, but the effect still couldn't meet the requirements. Now I'm trying to turn the picture into a bird's-eye view. Maybe only turning the camera to collect multi-angle data can really solve the problem. Thank you for your reply, which is of great help to me!