search

NazirNayal8 / RbA

Official code for RbA: Segmenting Unknown Regions Rejected by All (ICCV 2023)
https://kuis-ai.github.io/RbA/
MIT License
58 stars 6 forks source link

Question about train and finetune #16

Closed wyz22966 closed 1 month ago

wyz22966 commented 1 month ago

Hi, thanks to your effort on this fantastic work! I encountered some difficulties while reproducing the experiment. I am planning to train Swin-B on the Cityscapes dataset and finetune it on coco and mapillary, python train_net.py \ --config-file configs/cityscapes/semantic-segmentation/swin/single_decoder_layer/maskformer2_swin_base_IN21k_384_bs16_90k_1dl.yaml \ --num-gpus 1 SOLVER.IMS_PER_BATCH 8\ OUTPUT_DIR model_logs/swin_b_1dl/ I obtained model_final.pth using the aforementioned code. How should I proceed with fine-tuning it next?

NazirNayal8 commented 1 month ago

Hi @wyz22966 ,

Thank you for your interest in our work. We provide a config file that you can use to finetune on COCO and mapillary. You can find it here.

You can simply run where CONFIG is the path to the config file linked above:

python train_net.py --config-file CONFIG

This should exactly replicate the finetuning experiment. Make sure user-specific values, such as dataset paths, output paths ...etc are overwritten in the config file to match your setup.

I hope this was helpful. If you have anymore issues please write here in the issue again.

wyz22966 commented 1 month ago

Hello, thank you very much for your reply. python train_net.py --config-file ckpts/swin_b_1dl_rba_ood_map_coco/config.yaml --num-gpus 1 I encountered an error at the end of using the above code, but I still got the fine tuned 'model_final.pth' error I will put the PTH file into CKPTS/,and then evaluate it. python evaluate_ood.py --out_path results_test/only_finetune/ --models_folder ckpts --datasets_folder datasets/ --dataset_mode all --store_anomaly_scores the results are as follows:

1 finetune Thanks again for your help.