The dataset could not be loaded because the splits use different data file formats, which is not supported. Read more about the splits configuration. Click for more details.
Error code: FileFormatMismatchBetweenSplitsError
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.
YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
convnext-weakfix — data + kết quả + checkpoint
Replicate phương pháp v5_weakfix (hoangtuan) lên DINOv3 ConvNeXt-Tiny (dinov3_next_cnn).
Cùng data / kỹ thuật / hyperparam, identity-disjoint. Doc đầy đủ tại repo gốc:
/workspace/quangmanh/deepfake/doc/convnext_weakfix/.
Data (data/)
train_v5_combined_universal_kaggle_boost.csv— 54,000 ảnh (baseline).train_v5_weakfix.csv— v2: 121,884 ảnh (31,006 real / 90,878 fake, 42 method, faceswap 4,600).train_v5_weakfix_v3.csv— v3: 129,884 ảnh (faceswap 12,600).val_v5_combined_universal_kaggle_boost.csv— 6,000 ảnh val.test_balanced_fixed_zero_leakage.csv— test benchmark zero-leakage 2,354 ảnh (1:1 real/fake).v5_weakfix_dataset_summary.json,v5_weakfix_v3_dataset_summary.json— summary, identity_overlap_added=0.
CSV chỉ chứa đường dẫn ảnh + label/method/domain/identity (metadata), không kèm ảnh.
Kết quả (results/)
Eval theo protocol bf16 (giống v5): acc baseline 99.41% → v2 99.66% → v3 99.70%, faceswap v3 100% (27/27). Gồm eval_*.json/.npz, per_method_*.csv, faceswap_detail.csv, comparison.json, training_report.
Checkpoint (checkpoints/)
convnext_baseline.pt— v5-equivalent (3 epoch trên 54K).convnext_weakfix_v2.pt— method-balanced sampler, init từ baseline.convnext_weakfix_v3.pt— final (faceswap-focused, init từ v2).
Train: scripts/finetune_convnext_weakfix.py --stage {baseline|v2|v3} — Eval: scripts/eval_convnext_weakfix.py.
- Downloads last month
- 9