Datasets:
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
Error code: LockedDatasetTimeoutError
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.
model string | initialization string | epochs int64 | imgsz int64 | batch int64 | device int64 | workers int64 | seed int64 | ultralytics string | huggingface_hub string | data dict | test_policy string | scripts_sha256 dict |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
yolo11s.pt | COCO-pretrained; fresh optimizer and schedule, no resume from old 169-class model | 40 | 960 | 12 | 0 | 2 | 20,261,018 | 8.3.221 | 1.29.0 | {
"status": "complete",
"version": "v2",
"images": 31661,
"objects": 32464,
"classes": 167,
"images_by_split": {
"train": 25348,
"val": 3143,
"test": 3170
},
"image_bytes": 10254608839,
"manifest_sha256": "4285bef716e84fc9c3f64a01ea0c93ccfc0a1d24329d705c6491af322777f238",
"source_manifes... | Fixed test split evaluated only after training and validation model selection. | {
"kaggle_baseline.py": "b03559889a36fd96d948f38de2546e47aa55fe96bfde8b540d337c2decb7d8af",
"kaggle_prepare_cache.py": "2571643821118a32a0603dc450659bf6ae6683392960910ff7499fd802f0e610"
} |
TV Channel Logos
Current verified full-logo snapshot: 31,661 full-frame images, 32,464 complete-logo boxes, 167 appearance classes. Train / validation / test: 25,348 / 3,143 / 3,170 images. Updated 2026-10-19. Original image dimensions are preserved.
Layout
images/<prefix>/<sha256>.jpg is a content-addressed image pool. versions/v2/manifest.jsonl defines all current images, splits and normalized [x0,y0,x1,y1] full-logo boxes. versions/v2/class-catalog.json defines class IDs and names. UPLOAD_COMPLETE.json identifies the validated current snapshot.
Download a fixed repository revision, then run python materialize_yolo.py --root /path/to/snapshot --output /path/to/yolo. It verifies image SHA-256 and creates YOLO labels and links to the downloaded images. Keep the image pool available. The script does not train a model.
Sources and annotation
Legacy broadcast frames were converted from element annotations into whole-logo annotations. Later additions come from native live broadcast screenshots, with visual review of full frames and complete-logo boxes. Current metadata contains only whole-logo labels; the source element labels and obsolete releases are not included.
Visually distinct designs, including satellite-channel 4K variants, are separate classes. Overall aspect-ratio compression, scale, transparency and color alone do not create separate classes. Identical visible logos share a class. Display names use 睛彩, 卡酷少儿, and xx卫视_4K where applicable.
Known near-duplicate groups are kept in the same split; recent additions are train-only and screened against validation/test near-neighbors. Programme-level independence is not established. Three frames are negative examples. Legacy annotations combine automated checks and targeted visual review rather than exhaustive independent relabeling; errors may remain. source-validation.json records local validation, while UPLOAD_COMPLETE.json records the published snapshot.
Training
The previous 169-class YOLO11s model was trained for 40 epochs at input size 960. Those metrics do not evaluate this revised 167-class snapshot. The next authorized run uses YOLO11s, size 960, and a fresh 40-epoch schedule; it has not started at publication time.
Rights
Broadcast imagery and channel marks belong to their respective rights holders. No blanket image license is asserted. This repository retains the current snapshot; obsolete assets and prior commit history are intentionally retired after successful verification.
- Downloads last month
- 282