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/harbor/harbor.py", line 171, in _split_generators
raise DataFilesNotFoundError("No task.toml or instruction.md files found")
datasets.exceptions.DataFilesNotFoundError: No task.toml or instruction.md files found
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 68, 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.
TV Channel Logos
Upload status: COMPLETE. See UPLOAD_COMPLETE.json for transfer verification.
Whole channel-logo detection dataset. Current v2 covers mainly Chinese television channels; international expansion is a future goal, not current coverage.
30,503 full-frame images, 31,287 complete-logo boxes, 167 appearance classes. Train/validation/test: 24,190 / 3,143 / 3,170 images. Original resolutions are preserved. A YOLO11s baseline at input960 completed40 epochs; its fixed training data revision is f231ac92206fe51fc3d3e821ac209b972d3083f2.
Layout and use
images/<prefix>/<sha256>.jpg is an immutable image pool, shared across annotation versions. versions/v2/manifest.jsonl defines splits and normalized [x0,y0,x1,y1] whole-logo boxes. class-catalog.json maps class IDs. Images do not move when labels or splits change.
Download a completed repository revision, then run python materialize_yolo.py --root /path/to/snapshot --output /path/to/yolo-v2. This verifies every source image SHA-256 and creates a standard YOLO dataset with links to the local image pool. Keep that pool available. No training is started by this script.
Source and annotation policy
Images consist of 30,137 legacy training images, 228 initial broadcast screenshots and 138 additional broadcast screenshots. Legacy element annotations were reviewed and converted offline into complete-logo boxes. The inference task never requires element-combination rules.
Visually distinct logo designs are separate classes, including nine satellite-channel 4K variants. Global aspect-ratio distortion, scale, transparency or color alone do not create new classes. Identical visible marks share a class regardless of channel identity.
Quality limits
The local source release passed full file/pixel hash, box, count and known near-duplicate split-isolation checks. source-validation.json documents that LOCAL source audit, not remote upload completion. Legacy annotation quality combines automated checks and targeted visual review, not independent manual relabeling of every frame. Unknown errors may remain.
All current 167 classes occur in all three splits. Known near-duplicates are grouped; programme-level independence is not established. Three images are negative examples.
Rights
Broadcast imagery and channel marks belong to their respective rights holders. Public availability does not establish a blanket license for the underlying imagery; no MIT/Apache or other blanket image license is asserted here.
Display-name amendment (2026-10-18)
The nine satellite 4K classes (IDs158–166) now use xx卫视_4K. Only names changed; image bytes, numeric class IDs, boxes, and splits are unchanged. Existing trained weights retain their original names until their display mapping is separately updated. Original training snapshots and pinned recipes remain reproducible at their existing commit IDs.
Frame exclusion (2026-10-18)
Two user-rejected training frames (central station-ident artwork and pitch-side board logos) and all five boxes have been excluded from the active v2 manifest. Their now-empty classes167 and168 were removed; IDs0–166 remain unchanged. The immutable image pool retains historical source files for reproducibility, but only images referenced by the current manifest belong to this dataset version. The original 169-class training snapshot and existing model are unchanged.
- Downloads last month
- 222