【Hackathon + No.161】论文复现：CLRNet: Cross Layer Refinement Network for Lane Detection

[khan-yin]

training

implement environment:

• PaddlePaddle2.4.0
• V100*2

you need to change batch_size in configs/clrnet/_base_/clrnet_reader.yml when you use multiple GPUs

TrainReader:
  batch_size: 48 # 24 for one single GPU if use mutil GPUs please change it to 24*GPU_NUM

script:

export CUDA_VISIBLE_DEVICES=0,1
python -m paddle.distributed.launch --gpus 0,1 tools/train.py -c configs/clrnet/clr_resnet18_culane.yml

evaluation

script:

python tools/eval.py -c configs/clrnet/clr_resnet18_culane.yml -o weights=output/clr_resnet18_culane/model_final.pdparams output_eval=evaluation/ use_gpu=true

weights&logs

download link	code
BaiduDisk	fz71

.
|-- culane_r18.pth # torch pretrain model
|-- culane_r18.pdparams # export from torch by modified torch2paddle.py for CLRNet 
|-- model_final.pdparams # training model
|-- paddle_eval.log # use model_final.pdparams
|-- paddle_pretraineval.log # use culane_r18.pdparams
|-- paddle_training.log # use model_final.pdparams
|-- torch2paddle.py
`-- torch_eval.log # use torch pretrain model

backbone	mF1	F1@50	F1@75	weights
ResNet-18	55.39	79.56	62.83	model_final.pdparams
ResNet-18	55.57	79.58	62.79	culane_r18.pdparams
ResNet-18	55.32	79.58	62.55	culane_r18.pth

[paddle-bot]

Thanks for your contribution!

[LokeZhou]

这些.github的不用提交

[LokeZhou]

如果与clrnet无关的不用提交

[LokeZhou]

给一下训练、推理、评估的命令。还有数据集准备、下载链接

[LokeZhou]

同上

[LokeZhou]

确认是否需要pretrain_weights

[LokeZhou]

与clrnet无关的无需提交

[LokeZhou]

同上

[LokeZhou]

同上

[LokeZhou]

__init__里面的backbone等给默认值

[LokeZhou]

update最新develop

[LokeZhou]

同上

[LokeZhou]

预训练模型需要确认一下

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
7 out of 8 committers have signed the CLA.

✅ chunyuwei
✅ jerrywgz
✅ nemonameless
✅ ghostxsl
✅ zhiboniu
✅ LokeZhou
✅ thinkthinking
❌ kehanyin

---

kehanyin seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}