search

cfzd / Ultra-Fast-Lane-Detection

Ultra Fast Structure-aware Deep Lane Detection (ECCV 2020)
MIT License
1.82k stars 493 forks source link

loss 值振荡，不下降 #27

Closed gneworld closed 4 years ago

gneworld commented 4 years ago

你好，感谢你的工作，我在使用culane训练时，观察到的loss值分布如下，请问会是什么原因引起的，期待您的回复 QQ图片20200709094038

cfzd commented 4 years ago

@gneworld I wonder how long did the log last? If it is just a few epochs, the loss would not decrease a lot.

Also, I recommend you to use tensorboard to observe the losses and metrics. Our code has supported it.

gneworld commented 4 years ago

@cfzd I have trained on culane dataset for about 25 epochs with one 2080ti. As your advice in other issues, I changed the lr from 0.1 to 0.025, this is the latest trainind result 1594275427(1)

I don't know if it is right?

cfzd commented 4 years ago

@gneworld I think it looks like a good one. The iou is increasing and loss is decreasing. To verify the results, you can run

python test.py configs/culane.py --test_model path_to_culane_model.pth --test_work_dir ./tmp
gneworld commented 4 years ago

hi, cfzd After training 50 epochs, I got following test results, ....... res_normal 0.895078 res_crowd 0.674346 res_night 0.642249 res_noline 0.401987 res_shadow 0.661217 res_arrow 0.839021 res_hlight 0.603785 res_curve 0.595709 res_cross 0.0 0.6999230162150123

so is it seemed ok? But when I test the culane model on my own lane video, it works much worse than that in your demo video, I don't know what factors lead to this inconsistent result, looking forward to your reply

longzeyilang commented 4 years ago

@gneworld hi，have you get good result?