WZMIAOMIAO/deep-learning-for-image-processing · error · FileNotFoundError

VOCdevkit dose not in path:'{}'.

Error message

VOCdevkit dose not in path:'{}'.

What it means

train_multi_GPU.py checks that `<args.data_path>/VOCdevkit` exists before building VOCDataSet; if not, it raises FileNotFoundError. It ensures the distributed training has a valid VOC root.

Source

Thrown at pytorch_object_detection/faster_rcnn/train_multi_GPU.py:59

    device = torch.device(args.device)

    # 用来保存coco_info的文件
    results_file = "results{}.txt".format(datetime.datetime.now().strftime("%Y%m%d-%H%M%S"))

    # Data loading code
    print("Loading data")

    data_transform = {
        "train": transforms.Compose([transforms.ToTensor(),
                                     transforms.RandomHorizontalFlip(0.5)]),
        "val": transforms.Compose([transforms.ToTensor()])
    }

    VOC_root = args.data_path
    # check voc root
    if os.path.exists(os.path.join(VOC_root, "VOCdevkit")) is False:
        raise FileNotFoundError("VOCdevkit dose not in path:'{}'.".format(VOC_root))

    # load train data set
    # VOCdevkit -> VOC2012 -> ImageSets -> Main -> train.txt
    train_dataset = VOCDataSet(VOC_root, "2012", data_transform["train"], "train.txt")

    # load validation data set
    # VOCdevkit -> VOC2012 -> ImageSets -> Main -> val.txt
    val_dataset = VOCDataSet(VOC_root, "2012", data_transform["val"], "val.txt")

    print("Creating data loaders")
    if args.distributed:
        train_sampler = torch.utils.data.distributed.DistributedSampler(train_dataset)
        test_sampler = torch.utils.data.distributed.DistributedSampler(val_dataset)
    else:
        train_sampler = torch.utils.data.RandomSampler(train_dataset)
        test_sampler = torch.utils.data.SequentialSampler(val_dataset)

    if args.aspect_ratio_group_factor >= 0:

View on GitHub (pinned to 1ec3fe6f37)

Solutions

  1. Download/extract VOC2012 so `<data_path>/VOCdevkit` exists on every node used in the multi-GPU run
  2. Pass the parent of VOCdevkit via --data-path, not the VOCdevkit directory itself
  3. On multi-node setups, ensure the dataset is available at the same path on all ranks (shared filesystem or sync)

Example fix

# before
torchrun --nproc_per_node=2 train_multi_GPU.py --data-path /data/VOCdevkit
# after
torchrun --nproc_per_node=2 train_multi_GPU.py --data-path /data  # /data/VOCdevkit exists
Defensive patterns

Strategy: validation

Validate before calling

import os
data_path = args.data_path
assert os.path.exists(os.path.join(data_path, "VOCdevkit")), f"VOCdevkit missing under {data_path}"

Try / catch

try:
    launch_distributed_training(args)
except FileNotFoundError as e:
    if "VOCdevkit" in str(e):
        raise SystemExit(f"Dataset not found at {args.data_path}; extract VOC2012 there first.")

Prevention

When it happens

Trigger: Launching multi-GPU training with --data-path pointing to a directory lacking the VOCdevkit subfolder, or to VOCdevkit itself.

Common situations: Dataset not extracted on the training machine (common on multi-node setups where data exists only on one node); wrong --data-path value; symlink broken.

Related errors

AI-assisted analysis of WZMIAOMIAO/deep-learning-for-image-processing@1ec3fe6f37 (2026-08-30). Data as JSON: /api/errors/f0928fbd2179c7dc. Report an issue: GitHub.