danjacobellis/bdd100k_train
BDD100K Train This dataset was created for DeDelayed: Deleting Remote Inference Delay via On-Device Correction (CVPR 2026). Code is available at InterDigitalInc/dedelayed. The underlying data is derived from the BDD100K driving video dataset. It contains 69,800 training sequences of 15 frames each. Usage This dataset is the training split. Use it together with danjacobellis/bdd500_pl_f14 as follows: import datasets dataset = datasets.DatasetDict({ 'train':… See the full description on the dataset page: https://huggingface.co/datasets/danjacobellis/bdd100k_train.
| Author / organization | danjacobellis |
|---|---|
| License (as declared) | other |
| Downloads | 291 (as reported by Hugging Face) |
| Likes | 0 |
| Created | 20 August 2025 |
| Last updated at source | 3 April 2026 |
Popularity, last 2 days
291 now (+5 since 2026-10-16). Counts as reported by the source, snapshotted daily by Silytics.
More from danjacobellis
- danjacobellis/LSDIRDataset
- danjacobellis/LSDIR_rawDataset
- danjacobellis/chexpertDataset
- danjacobellis/musdb18HQDataset
- danjacobellis/audioset_opus_24kbpsDataset
- danjacobellis/vimeo90k_tripletDataset
Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.