VGGFace2-HQ / README.md
ProgramComputer's picture
Revise access terms and disable viewer
5002426 verified
|
Raw History Blame Contribute Delete
2.5 kB
---
task_categories:
- image-classification
license: cc-by-nc-4.0
viewer: false
extra_gated_prompt: >-
This dataset contains images or recordings of real, identifiable people that
third parties collected from the web. Copyright in the media (images, audio and
video) stays with its original owners, and access here is not a licence to it.
Some parts, such as the original publishers' annotations, carry their own
licences; these conditions do not limit any rights those licences give you.
Recognition evaluation on the dataset itself is allowed. Read the dataset card
before use.
extra_gated_fields:
I will use the media only for non-commercial research: checkbox
I will not redistribute the media: checkbox
I will not use the dataset to identify, track or monitor people outside it: checkbox
I will delete affected media if told of a removal request: checkbox
---
https://github.com/NNNNAI/VGGFace2-HQ
## What this repository contains
- `train/VGGFac01.zip` to `train/VGGFac10.zip`: the VGGFace2-HQ project's
training release, downloaded from its Google Drive and re-packaged into ten
ZIP files.
- `test/test.zip`: not part of the project's release. It was generated for this
repository from the VGGFace2 test set with the project's own scripts: FFHQ
alignment (`scripts/crop_align_vggface2_FFHQalign.py`), then GFPGAN
restoration (`scripts/inference_gfpgan_forvggface2.py` with
`GFPGANCleanv1-NoCE-C2.pth`). It holds 61,841 restored images.
## Removal requests
If you appear in this dataset, or hold rights in any of its media, and want it
removed, open a discussion in this repository's Community tab with the file
paths or IDs involved.
## Bounded streaming
`datasets==5.0.0` does not execute loading scripts from the Hub. Clone code and
metadata without downloading Git LFS objects, then import the loader directly:
```shell
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/datasets/ProgramComputer/VGGFace2-HQ
```
Pin both data revisions when constructing the stream:
```python
from VGGFace2_HQ import load_streaming
dataset = load_streaming(
"train",
revision="ee41c3e64981063524744286b468e739190cfbed",
raw_revision="ad5f6b5a5f560621fd7efb9b79c956d27d427a08",
cache_dir="/bounded/cache",
scratch_dir="/bounded/scratch",
)
first_record = next(iter(dataset))
```
ZIP shards use bounded HTTP-range reads. A full-shard scratch fallback is
disabled unless `allow_full_shard_fallback=True` is passed with explicit byte
limits.