Hugging Face's logo

Spaces:
bayat10
/
media-authenticity
Running on Zero

App Files Community
media-authenticity / models

Commit History

update
fb97e0d
Running

cryptomathematician commited on

update
7afc59e

cryptomathematician commited on

update
f67ba3e

cryptomathematician commited on

update
67b0d80

cryptomathematician commited on

update
27e0d27

cryptomathematician commited on

updates
f300fa8

cryptomathematician commited on

updates
1c04f54

cryptomathematician commited on

updates
89f4bff

cryptomathematician commited on

updates
5fa308a

cryptomathematician commited on

updates
53b0564

cryptomathematician commited on

update
cb76d12

cryptomathematician commited on

Using training pipeline
af1c911

cryptomathematician commited on

added frequency
a1c52af

cryptomathematician commited on

Track NPR deepfake detection models
e485d53

cryptomathematician commited on

added npr repository
6c86f3f

cryptomathematician commited on

fixed additional corruptions
1fa4b15

cryptomathematician commited on

fixed additional corruption
d9e98ee

cryptomathematician commited on

fixed additional LFM corruptions
cff0fd8

cryptomathematician commited on

fixed the indentation error
7f3bd87

cryptomathematician commited on

gated fusion done
4366522

cryptomathematician commited on

made the necessary changes to ungated fusion
68153a6

cryptomathematician commited on

added the ungated fusion training scripts
938ed32

cryptomathematician commited on

ai video detector
0fdd219

cryptomathematician commited on

made sure if any frame scores fake then the video is fake
552e317

cryptomathematician commited on

fixed list undefined error
0f800d4

cryptomathematician commited on

frame score added
19a4ecd

cryptomathematician commited on

made sure the frame sampling is evenly distributed
db6f866

cryptomathematician commited on

added different video detector
634e783

cryptomathematician commited on

fixed bugged identified
58343f8

cryptomathematician commited on

Fix: center-crop non-square frames to square before resize (match DeCoF dataset prep)
1357406

cryptomathematician commited on

Unify DeCoF frame extraction: both predict() and predict_from_video_path() use first-32-frames protocol
65010fa

cryptomathematician commited on

Fix: extract frames from first 32 frames only (matching DeCoF README protocol)
98ace57

cryptomathematician commited on

Add DeCoF debug logging for logits/feature stats, use plain threshold
6332a95

cryptomathematician commited on

Fix DeCoF preprocessing to match original repo (ImageNet norm, no L2 normalization)
1278cca

cryptomathematician commited on

Fix DeCoF: deterministic frame sampling, HF checkpoint download, format detection
ff31fd7

cryptomathematician commited on

deterministic video prediction
637de21

cryptomathematician commited on

fixed the error with video prediction
fe19e5a

cryptomathematician commited on

Gitignore binary files (png, pth) and remove from tracking
7c709da

cryptomathematician commited on

Add DeCoF video deepfake detector, update gend detector and app, remove cross_efficient_vit models
72f1694

cryptomathematician commited on

Use ungated_fusion_fold1 weights for image prediction
417b891

cryptomathematician commited on

Use fold 2 weights for UngatedFusion detector
7e3c78c

cryptomathematician commited on

Fix UngatedFusion loading: use fold checkpoint for LFM weights, no separate lfm_bs_model.pth needed
1c5e884

cryptomathematician commited on

Lower GenD default threshold to 0.5 for better classification balance
4614d04

cryptomathematician commited on

Fix GenD checkpoint namespace by adding missing backbone.model prefix
9cba24a

cryptomathematician commited on

Fix state_dict key mismatch: strip extra .model from checkpoint backbone keys
bfdd043

cryptomathematician commited on

Fix state_dict key mismatch: rename backbone.layer to backbone.model.layer before load_state_dict
c9f2c53

cryptomathematician commited on

Restore load_state_dict after initialize_model to load trained classifier weights
bf7b739

cryptomathematician commited on

Fix GenD loading: defer feature_extractor/head init to initialize_model(), avoid nested from_pretrained
ed6d0c7

cryptomathematician commited on

Add device_map=None to GenD.from_pretrained to disable accelerate meta initialization
9cb37b5

cryptomathematician commited on

Refactor GenD loader: use GenD.from_pretrained with low_cpu_mem_usage=False, cache processor
f1bda16

cryptomathematician commited on