Before diving into this, please make sure you followed the instructions to prepare datasets in DATASET.md
Execution is based on config files
Some models' ImageNet pre-trained weights need to be manually downloaded, refer to this table.
python main_landet.py --train \
--config=<config file path> \
--mixed-precision # Optional, enable mixed precision \
--cfg-options=<overwrite cfg dict> # Optional
Your <overwrite cfg dict>
is used to manually override config file options in commandline so you don't have to modify config file each time. It should look like this (the quotation marks are necessary!): "train.batch_size=8 train.workers=4 model.lane_classifier_cfg.dropout=0.1"
Some options can be used by shortcuts, such as --batch-size
will set both train.batch_size
and test.batch_size
, for more info:
python main_landet.py --help
Example shells are provided in tools/shells.
We support multi-GPU training with Distributed Data Parallel (DDP):
python -m torch.distributed.launch --nproc_per_node=<number of GPU per-node> --use_env main_landet.py <your normal args>
With DDP, batch size and number of workers are per-GPU. Do not forget to set device args like world_size
in your config.
Important Notice: Do not simoutanously run multiple evaluation on CULane, since the eval use the same pytorch-auto-drive/output cache directory, the results could be overwritten! Same goes for LLAMAS!
- Predict lane lines:
python main_landet.py --test \ # Or --val for validation
--config=<config file path> \
--mixed-precision # Optional, enable mixed precision \
--cfg-options=<overwrite cfg dict> # Optional
To test a downloaded pt file, try add --checkpoint=<pt file path>
.
Note that LLAMAS doesn't have test set labels.
- Test with official scripts on
<my_dataset>
:
./autotest_<my_dataset>.sh <exp_name> <mode> <save_dir>
<mode>
includes test
and val
.
<save_dir>
and <exp_name>
are recommended to set the same as in config file, so detail evaluation results will be saved to <save_dir>/<exp_name>/
Overall result will be saved to log.txt
.
Training contains online fast validations by using val_num_steps
and the best model is saved, but we find that the best checkpoint is usually the last, so probably no need for validations. For log details you can checkout tensorboard.
To validate a trained model on mean IoU, you can use either mixed-precision or fp32 for any model trained with/without mixed-precision:
python main_landet.py --valfast \
--config=<config file path> \
--mixed-precision # Optional, enable mixed precision \
--cfg-options=<overwrite cfg dict> # Optional