The dataset viewer is not available for this subset.
Exception: SplitsNotFoundError
Message: The split names could not be parsed from the dataset config.
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
for split_generator in builder._split_generators(
~~~~~~~~~~~~~~~~~~~~~~~~~^
StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 81, in _split_generators
first_examples = list(islice(pipeline, self.NUM_EXAMPLES_FOR_FEATURES_INFERENCE))
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 32, in _get_pipeline_from_tar
fs: fsspec.AbstractFileSystem = fsspec.filesystem("memory")
~~~~~~~~~~~~~~~~~^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/fsspec/registry.py", line 302, in filesystem
cls = get_filesystem_class(protocol)
File "/usr/local/lib/python3.14/site-packages/fsspec/registry.py", line 239, in get_filesystem_class
raise ValueError(f"Protocol not known: {protocol}")
ValueError: Protocol not known: memory
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 71, in compute_split_names_from_streaming_response
for split in get_dataset_split_names(
~~~~~~~~~~~~~~~~~~~~~~~^
path=dataset,
^^^^^^^^^^^^^
config_name=config,
^^^^^^^^^^^^^^^^^^^
token=hf_token,
^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
info = get_dataset_config_info(
path,
...<6 lines>...
**config_kwargs,
)
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Market-1501 Gray Background
This dataset is a gray-background derivative of the Market-1501 person
re-identification dataset. Person masks were predicted with
YOLO26x-seg.
For images with an accepted person mask, foreground color is retained and the
background is replaced with neutral gray (RGB 127, 127, 127).
The dataset was generated with one fixed-confidence segmentation pass. Images for which YOLO26x-seg did not return an accepted person mask were copied unchanged from the original dataset. No lower-confidence recovery pass was performed.
Archive contents
The repository contains:
Market-1501-gray.tar.gz
After extraction:
Market-1501-gray/
├── bounding_box_train/
├── bounding_box_test/
├── query/
├── gt_bbox/
├── gt_query/
└── readme.txt
| Item | Count |
|---|---|
| Total images | 61,295 |
| Segmented images | 59,942 |
| Original-image fallbacks | 1,353 |
| Non-image metadata files | 6,741 |
| Total files | 68,036 |
The complete original Market-1501 directory structure, relative paths, filenames, junk images, distractors, ground-truth bounding boxes, and evaluation metadata are preserved.
Generation command
The initial segmentation pass was run from the BoxMOT repository:
caffeinate -i -s uv run python -m tools.create_market1501_person_masks \
--source Market-1501-v15.09.15 \
--output Market-1501-gray \
--model https://github.com/ultralytics/assets/releases/download/v8.4.0/yolo26x-seg.pt \
--device mps \
--batch-size 16 \
--conf 0.10
Initial-pass result:
Masking Market-1501: 100%|████████████████████| 61295/61295
written=59942, missing_masks=1353
The 1,353 paths reported without masks were then copied byte-for-byte from
Market-1501-v15.09.15 into the corresponding locations under
Market-1501-gray. They were not passed through segmentation again.
Mask processing
For each accepted prediction:
- Inference was restricted to the YOLO person class.
- When several people were detected, the primary mask was selected using confidence, mask area, and proximity to the crop center.
- The mask was resized to the source image's exact width and height.
- The mask was binarized at a threshold of
0.5. - Pixels outside the mask were set to gray 127.
Processing used:
| Setting | Value |
|---|---|
| Model | YOLO26x-seg |
| Inference size | 640 |
| Detection confidence | 0.10 |
| NMS IoU | 0.70 |
| Mask threshold | 0.50 |
| Background gray | 127 |
| Mask dilation | 0 |
| Edge feathering | 0 |
imgsz=640 controls model inference resolution only. Every generated image
has the same pixel dimensions as its source image. Generated JPEGs were
encoded at quality 95, so their byte sizes and compressed pixel values differ
from the originals. Fallback images are exact copies of their source files.
Validation and integrity
- All 61,295 original image paths are present.
- All 68,036 original dataset file paths are present.
- All 1,353 fallback images are byte-identical to their originals.
- No low-confidence second segmentation pass was used.
Archive integrity:
Filename: Market-1501-gray.tar.gz
Size: 162,056,902 bytes
SHA-256: b6909a8cd4e9d2104acb7455a8efb62c95cad1f0db9e65294886511922ff2fae
Limitations
- The person masks are model predictions, not human annotations.
- Some accepted masks can contain segmentation errors, especially for small, occluded, or partial-body crops.
- The 1,353 fallback images retain their original backgrounds, so background treatment is not uniform across the dataset.
- Gray backgrounds change the visual domain and may affect compatibility with models trained on the original Market-1501 images.
- PID
0000gallery distractors remain evaluation negatives, while PID-1junk images are ignored by standard BoxMOT Market-1501 loading.
Source, citation, and terms
Market-1501 was introduced in:
Liang Zheng, Liyue Shen, Lu Tian, Shengjin Wang, Jingdong Wang, and Qi Tian. Scalable Person Re-identification: A Benchmark. ICCV 2015.
The original Market-1501 README states that the dataset is for research use and requests that it not be redistributed or used commercially. Users are responsible for obtaining Market-1501 legitimately and complying with the terms of the source dataset and YOLO26x-seg.
- Downloads last month
- 56