dongkwonjin / RVLD

Official code for "Recursive Video Lane Detection", ICCV 2023
Apache License 2.0
77 stars 9 forks source link

Which model do we use if we wish to train a model which re-creates paper results? #12

Open aj96 opened 10 months ago

aj96 commented 10 months ago

I would like to train a model which achieves your SOTA result of 0.825 F1@0.5 on OpenLane-V dataset. Which model/folder should I use for doing this? I assume I should use this one Modeling/OpenLane-V/PLD/code? Thank you for the amazing work.

Edit: Never mind. I should have more carefully read the paper. It seems you train ILD first and then train PLD. It also seems that you train ILD in two separate stages as well, the probability map stage and the coefficient map stage, so the model requires training 3 stages in total? Did you try training the model end-to-end? Was there a significant benefit to training these three stages separately? Thank you.

lwx980315 commented 9 months ago

I also have this question!

Mrzhang-meng commented 6 months ago

@aj96 您好,我想问下你在执行的过程中是否按照三个过程依次执行,并且在每个阶段都进行train,test,和eval,然后pld的结果就是论文得到的结果呢,谢谢。

lwx980315 commented 6 months ago

@aj96 您好,我想问下你在执行的过程中是否按照三个过程依次执行,并且在每个阶段都进行train,test,和eval,然后pld的结果就是论文得到的结果呢,谢谢。

你好!我已经完整复现了这篇工作,并理解其中每一处细节。作者给到的每个数据集中的3个文件夹是需要按顺序训练的,真正出实验结果写在论文中的是PLD,前面两个也可以测试,结果供自己看就好。

dongkwonjin commented 6 months ago

@aj96 @lwx980315 @Mrzhang-meng End-to-end approach is also possible. But, we separately trained each model for training stability. With this scheme, it is also more efficient to develop and optimize an overall algorithm, as only the desired parts, such as encoder, decoder, and motion estimator, can be trained without unnecessary retraining of the entire model. Thank you for your attention and we are sorry for the late reply. If you have any questions, please feel free to ask at any time.

Mrzhang-meng commented 6 months ago

十分感谢您的回复,我会才采取您宝贵的意见,谢谢!

---Original--- From: @.> Date: Thu, May 16, 2024 17:23 PM To: @.>; Cc: @.**@.>; Subject: Re: [dongkwonjin/RVLD] Which model do we use if we wish to train amodel which re-creates paper results? (Issue #12)

@aj96 您好,我想问下你在执行的过程中是否按照三个过程依次执行,并且在每个阶段都进行train,test,和eval,然后pld的结果就是论文得到的结果呢,谢谢。

你好!我已经完整复现了这篇工作,并理解其中每一处细节。作者给到的每个数据集中的3个文件夹是需要按顺序训练的,真正出实验结果写在论文中的是PLD,前面两个也可以测试,结果供自己看就好。

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

Mrzhang-meng commented 6 months ago

十分感谢您的回复,我会参考您宝贵的意见!

---Original--- From: "Dongkwon @.> Date: Thu, May 16, 2024 18:03 PM To: @.>; Cc: @.**@.>; Subject: Re: [dongkwonjin/RVLD] Which model do we use if we wish to train amodel which re-creates paper results? (Issue #12)

@aj96 @lwx980315 @Mrzhang-meng End-to-end approach is also possible. But, we separately trained for training stability. With this scheme, it is also more efficient to develop and optimize an overall algorithm, as only the desired parts, such as encoder, decoder, and motion estimator, can be trained without unnecessary retraining of the entire model. Thank you for your attention and we are sorry for the late reply. If you have any questions, please feel free to ask at any time.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

Mrzhang-meng commented 6 months ago

@dongkwonjin 作者您好,我在训练过程中有一些问题想请教您一下,在执行命令python main.py --run_mode test --pre_dir ROOT/preprocessed/DATASET_NAME/ --dataset_dir /where/is/your/dataset/path, readme中介绍的是对模型进行评估,意思是命令中的test就包含测试结果和评估结果吗,其次,在configs文件中self.model有四种模式,我修改为eval和test,得到的结果是一样样的,这是为什么?最后,python main.y --run_mode test_paper --pre_dir ROOT/preprocessed/DATASET_NAME/ --paper_weight_dir ROOT/pretrained/DATASET_NAME/ --dataset_dir /where/is/your/dataset/path这个命令是对你保存的权重文件进行测试吗,但是我在运行的时候, 显示No such file or directory: '/home/data1/zsm/RVLD/Modeling/VIL-100/PLD/output/test/results/iou/txt/results_txt/results/4_Road026_Trim001_frames.txt',这是需要在哪里产生的呢。在此,非常希望得到您的回复,谢谢!

dongkwonjin commented 6 months ago

1) 'test' mode includes both inference and evaluation processes, whereas 'eval' includes the evaluation only.

2) Because the evaluation is performed using prediction results obtained in the inference ('test' mode)

3) Please recheck your data path. After inferencing the model, the text files are generated. If you find and know about the path of them, you can fix the error by matching the path.

Thank you for your attention :)