训练自己的数据集

[HarryMark927]

作者你好！抱歉打扰
想请教一下如果我想在自己的数据集上进行训练和测试，我需要在哪几个文件上进行修改呢？

[MendelXu]

1. Add a script to define your dataset in san/data/datasets. You can refer to https://github.com/MendelXu/SAN/blob/main/san/data/datasets/register_voc.py. Remember to import it in san/data/datasets/__init__.py.
2. Train the model with python train_net.py --eval-only --config-file <CONFIG_FILE> --num-gpus <NUM_GPU> OUTPUT_DIR <OUTPUT_PATH> MODEL.WEIGHTS <TRAINED_MODEL_PATH> MODEL.SAN.NUM_CLASSES [NUM_OF_CLASS].

Hint To make the training process work properly, please check the batch data before training for a long time (You can add a breakpoint Here .

[HarryMark927]

sorry to bother you again.
I run the command "python train_net.py --config-file ./configs/san_clip_vit_res4_coco.yaml --num-gpus 1 OUTPUT_DIR ./log"

but error with " Traceback (most recent call last):
File "/data2/borong/SAN/train_net.py", line 280, in
launch(
File "/data2/borong_anaconda/envs/san/lib/python3.9/site-packages/detectron2/engine/launch.py", line 82, in launch
main_func(*args)
File "/data2/borong/SAN/train_net.py", line 271, in main
trainer = Trainer(cfg)
File "/data2/borong_anaconda/envs/san/lib/python3.9/site-packages/detectron2/engine/defaults.py", line 378, in init
data_loader = self.build_train_loader(cfg)
File "/data2/borong/SAN/train_net.py", line 94, in build_train_loader
return build_detection_train_loader(cfg, mapper=mapper)
File "/data2/borong_anaconda/envs/san/lib/python3.9/site-packages/detectron2/config/config.py", line 207, in wrapped
explicit_args = _get_args_from_config(from_config, *args, **kwargs)
File "/data2/borong_anaconda/envs/san/lib/python3.9/site-packages/detectron2/config/config.py", line 245, in _get_args_from_config
ret = from_config_func(*args, **kwargs)
File "/data2/borong/SAN/san/data/build.py", line 161, in _train_loader_from_config
dataset = get_detection_dataset_dicts(
File "/data2/borong/SAN/san/data/build.py", line 124, in get_detection_dataset_dicts
dataset_dicts = [
File "/data2/borong/SAN/san/data/build.py", line 125, in
wrap_metas(DatasetCatalog.get(dataset_name), dataset_name=dataset_name)
File "/data2/borong_anaconda/envs/san/lib/python3.9/site-packages/detectron2/data/catalog.py", line 58, in get
return f()
File "/data2/borong/SAN/san/data/datasets/register_coco_stuff_164k.py", line 212, in
lambda x=image_dir, y=gt_dir: load_sem_seg(
File "/data2/borong_anaconda/envs/san/lib/python3.9/site-packages/detectron2/data/datasets/coco.py", line 266, in load_sem_seg
(os.path.join(image_root, f) for f in PathManager.ls(image_root) if f.endswith(image_ext)),
File "/data2/borong_anaconda/envs/san/lib/python3.9/site-packages/iopath/common/file_io.py", line 1294, in ls
return self.__get_path_handler(path)._ls(path, **kwargs)
File "/data2/borong_anaconda/envs/san/lib/python3.9/site-packages/iopath/common/file_io.py", line 714, in _ls
return os.listdir(self._get_path_with_cwd(path))
FileNotFoundError: [Errno 2] No such file or directory: 'datasets/coco/train2017'
"
I checked the code but I did not find how to replace the coco dataset with my dataset that I want to train, and I have already prepared my dataset as you said like voc. And my propose is to train my own dataset in your model and later test it.

[MendelXu]

python train_net.py --eval-only --config-file <CONFIG_FILE> --num-gpus <NUM_GPU> OUTPUT_DIR <OUTPUT_PATH> MODEL.WEIGHTS <TRAINED_MODEL_PATH> MODEL.SAN.NUM_CLASSES [NUM_OF_CLASS] DATASETS.TRAIN <YOUR_DATASET_NAME>