bug: fixed path of coco 128

[znsoftm]

Traceback (most recent call last): File "train.py", line 238, in main() File "train.py", line 212, in main train_loader, val_loader = get_dataloader(opt, no_aug=no_aug) File "/home/znsoft/yolox-pytorch/data/coco_dataset.py", line 18, in get_dataloader train_dataset = COCODataset(data_dir=opt.data_dir, File "/home/znsoft/yolox-pytorch/data/datasets/coco.py", line 49, in init assert os.path.isfile(json_file), 'cannot find {}'.format(json_file) AssertionError: cannot find /data/dataset/coco_dataset/annotations/instances_train2017.json

I have changed the coco path to /mnt/d/coco_obj , it still complains a wrong path.

[zhangming8]

You still using my dataset path. Maybe, you not modify 'dataset_path' to "/mnt/d/coco_obj" in "train.sh"

[znsoftm]

如果在train.sh中可以指定数据集位置，那是不是就不要在config.py中再加了？

还有上次说的，建议类别从数据中读取，不要写死在config.py中，毕竟正常训练的数据肯定不是coco原生的，否则训练的意义不大

[zhangming8]

嗯，在*.sh中传入的参数会替换掉opt.xxx里面的参数