Comparison of Fine-Tuned Networks on Generalization for Face Spoofing Detection
Bibliographic record
Abstract
Spoofing is a primary security concern for all the organizations and researchers across the globe. Security can be achieved through different mediums; authentication is one such important medium. Biometric Authentication is considered as an important and strong form that's difficult to break. Biometric authentication mainly includes two mechanisms, viz. Physiological and Behavioral, Physiological traits include the face, fingerprint, retina, iris, palm geometry, etc. Face Recognition has many application areas due to its ease of implementation, and they can be easily fooled or spoofed, termed as Face Spoofing Attack. Face spoofing attacks are viz. 2D and 3D attacks, 2D Attacks include Fake photo, Warped photos, Video display and 3D attacks performed using 3D masks. Deep learning methods have proved beneficial for detecting spoofing attacks; these methods use fine-tuned and pre-trained models. The paper compares the proposed fine-tuned VGG16 and RESNET-50 architectures and their generalization performance of Face Spoofing Detection. The 3D MAD and NUAA Imposter Dataset are used to validate the performance for two color spaces viz. RGB and YCBCR; the results are obtained for both color spaces. RGB color space is related to human visual system but it's not invariant to illumination on the other hand YCBCR separates chrominance and luminance part which makes it illumination invariant and face recognition systems have reflectance issue. Cross-dataset evaluation is an important metric for face liveness detection. The paper presents cross dataset results on the above datasets with the lowest HTER of 18%. The fine-tuned VGG-16 architecture gives the best values for cross-dataset evaluation when trained on 3D MAD and tested for NUAA imposter dataset and same is true for RESNET-50 architecture.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.004 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".