ForensicNet: Lightweight Attention-Enhanced MobileNetV2 for Automated Face Identification

Lata B T; Savitha N J

REVIEW 4 major objections 5 minor 23 references

ForensicNet: Lightweight Attention-Enhanced MobileNetV2 for Automated Face Identification

T0 review · 4 major / 5 minor · reviewed 2026-08-02 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read ForensicNet claims 92.4% face-identification accuracy at 2.1 GFLOPs by adding CBAM attention and two-phase transfer learning to MobileNetV2.

desk verdict Standard MobileNetV2+CBAM recipe with a plausible 92.4% on a self-curated 68-class set, but the dataset description does not add up and the FLOPs count looks off; the qualitative direction is fine, the headline number is unverifiable. read the letter →

arxiv 2607.16273 v1 pith:XTCXKG4X submitted 2026-07-08 cs.CV cs.AI

Savitha N J , Lata B T This is my paper · ORCID

classification cs.CVcs.AI

keywords forensicfaceidentificationMobileNetV2CBAMattentiontransferlearningrecognitionsurveillancelightweightCNNGFLOPs

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that a lightweight MobileNetV2 face classifier can be made more accurate for forensic-style identification by adding Convolutional Block Attention Modules (CBAM) and using a two-phase transfer-learning schedule that freezes early layers before gradually unfreezing them. On a self-curated dataset of 15,000 facial images across 68 identities drawn from LFW and SCFace, the authors report 92.4% accuracy, 90.8% precision, and 89.5% recall at 2.1 GFLOPs per inference, surpassing AlexNet, ResNet-50, and vanilla MobileNetV2 under the same protocol. If these numbers hold, the significance is practical: a small, fast model could be deployed for real-time or edge surveillance without sacrificing identification accuracy. The contribution is an engineering recipe — attention injection plus an adaptive unfreezing schedule — rather than a new theoretical result.

What carries the argument

The central mechanism is a Convolutional Block Attention Module attached to the MobileNetV2 feature extractor, combined with a binary freezing mask m(l) that selects which layers stay trainable during transfer. Phase 1 freezes the backbone's early layers and trains only the attention modules and classification head; Phase 2 progressively unfreezes deeper layers at a reduced learning rate. The mask and update rule carry the argument: preserving generic visual features first, then adapting high-level representations to the forensic domain. The paper's equations explicitly describe channel attention with global average and max pooling; the text refers to CBAM's channel and spatial attention.

What would settle it

Attempt to reconstruct the dataset by listing the 68 identities and the per-source image counts; if the SCFace images cannot be assigned to those classes, or if retraining on a faithful reconstruction yields accuracy significantly below 92.4%, the central claim fails.

Watch

Extended reading notes

Core claim

The central claim is that CBAM attention and a two-phase transfer-learning schedule each add a measurable improvement on top of MobileNetV2. The paper's ablation shows vanilla MobileNetV2 at 89.1%, +CBAM at 91.3%, and +two-phase training at 92.4%, with precision 90.8%, recall 89.5%, and F1 90.1%. The authors attribute this to attention-refined feature maps that suppress background and occlusion, and to phase 1 freezing the early backbone while training the classification head and attention, followed by progressive unfreezing of deeper layers at a reduced learning rate. They report 2.1 GFLOPs per inference, below ResNet-50 (3.8), EfficientNet-B0 (3.9), ViT-Small (4.6), GhostNet (2.5), and van

Load-bearing premise

The load-bearing premise is that the Section III.A dataset — 15,000 images across 68 identity classes from LFW and SCFace — is assembled as described, with no stated mapping for the 4,200 SCFace images into those classes; if that construction is wrong, unrepresentative, or unreproducible, the 92.4% result cannot be verified.

Editorial extensions

If this is right

On the curated 68-identity benchmark, ForensicNet reports 92.4% accuracy, 3.3 points above vanilla MobileNetV2 and 3.7 points above ResNet-50.
The ablation supports that both ingredients matter: CBAM alone adds 2.2 points, and the two-phase schedule adds another 1.1 points.
At 2.1 GFLOPs per inference, the model is the cheapest among the compared architectures (ResNet-50 3.8, ViT-Small 4.6, EfficientNet-B0 3.9, GhostNet 2.5, MobileNetV2 2.3).
Higher recall (89.5%) than the baselines implies fewer missed true-positive matches, which the authors connect to surveillance utility.
The paper presents the combination as suitable for real-time and edge-based forensic surveillance because of its low computational cost.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The 68-class closed-set protocol sidesteps the open-set nature of real forensics; a natural extension is to hold out unknown identities and measure rejection and false-accept rates.
Independent reconstruction of the exact identity-to-source mapping would settle whether the dataset arithmetic holds: the paper does not state how the 4,200 SCFace images map onto the 68 identity classes, even though SCFace's public subject structure is much larger.
As written, the CBAM equation shows only channel attention; if spatial attention is intended, ablating the channel and spatial branches separately would clarify which one drives the reported gain.
A latency benchmark in frames per second on a phone-class device would make the 'real-time' claim concrete, since GFLOPs alone do not guarantee throughput.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Read this one if you want an example of a modest engineering result whose real problem is evaluation transparency. The method is MobileNetV2 + CBAM + progressive unfreezing; nothing surprising. The ablation is monotone (89.1 -> 91.3 -> 92.4), which matches the wider literature, so the qualitative direction is probably right.

To their credit, they include comparisons against EfficientNet, GhostNet, and ViT-Small under identical splits (Table III), and the conclusion is honest about limitations — real-world performance may vary. They also show Grad-CAM maps. That is more than many papers at this level do.

The load-bearing problem is Section III.A. They say the curated set has 15,000 images, 68 identities, 10,800 LFW + 4,200 SCFace. SCFace has roughly 130 subjects, so 4,200 SCFace images cannot sit inside 68 identity classes without either discarding most subjects or merging distinct identities — no explanation is given. That makes the 92.4% headline unreproducible. Add to that: no error bars, single-run point estimates in Tables I–III; the reported 2.1 GFLOPs for ForensicNet vs 2.3 for MobileNetV2 is anomalous, and standard counts for MobileNetV2 at 224x224 are around 0.3–0.6 GFLOPs. Also no margin-based face recognition baseline (ArcFace/SphereFace) despite citing those as SOTA in the intro. No code or exact split is shipped.

These are addressable. If the authors release the exact identity list and split, and correct the FLOPs count, the central claim can be checked. I would not reject the idea outright; it is a standard recipe applied to an interesting domain, and the ablation supports the design.

Who benefits? Anyone tracking lightweight face identification on edge devices. It deserves a serious referee — conditional on the authors fixing the dataset description and rerunning with seeds and error bars. It should not be desk-rejected, but it should not be accepted as-is.

Referee Report

4 major / 5 minor

Summary. The paper proposes ForensicNet, a lightweight deep network for automated forensic face identification under surveillance conditions. The architecture combines a MobileNetV2 backbone with CBAM attention modules and a two-phase transfer-learning strategy with adaptive layer unfreezing. The model is evaluated on a self-curated dataset of 15,000 face images spanning 68 identity classes, built from LFW and SCFace with an 80/10/10 train/validation/test split. The headline results are 92.4% accuracy, 90.8% precision, 89.5% recall, 90.1% F1-score, and 2.1 GFLOPs per inference, claimed to outperform AlexNet, ResNet-50, and vanilla MobileNetV2 (Tables I and III). The ablation study reports monotonic gains: MobileNetV2 baseline 89.1%, +CBAM 91.3%, +two-phase training 92.4% (Table II). The central claim is empirical rather than theoretical.

Significance. If the reported results are correct, this is a modest but practically relevant engineering contribution: a standard lightweight backbone with CBAM-style attention and a two-phase fine-tuning schedule yields roughly 3.3 accuracy points over vanilla MobileNetV2 on a closed-set 68-identity task, at a claimed FLOPs level suitable for edge deployment. The architecture is conventional, the ablation direction is plausible, and the paper builds on two publicly available datasets. However, the paper provides no code, no exact class/split list, no error bars, and no FLOPs counting details, and the dataset description in Section III.A is internally inconsistent as written. As a result, the central quantitative claims cannot currently be independently checked. A revised version with a transparent benchmark definition and significance-aware reporting could make this a valid practical validation study.

major comments (4)

[Section III.A and Tables I-III] The dataset description cannot be literally true. SCFace contains roughly 130 subjects and about 4,160 images; assigning 4,200 SCFace images to only 68 identity classes requires either discarding most SCFace subjects (contradicting the count) or merging different subjects into one label (contradicting the identity-class claim). The sentence 'Only identity classes containing at least 15 facial images were selected' does not explain how LFW and SCFace identities are matched, which subset was kept, or how 10,800 LFW images and 4,200 SCFace images are distributed across 68 classes. The exact class-to-subject mapping, per-class image counts, and the train/validation/test identity split are not given. Since every accuracy, precision, recall, and F1 value in Tables I-III is measured on this unreleased benchmark, the 92.4% headline cannot be independently verified. Please provide the full curate
[Table III and Section III.B (GFLOPs)] ForensicNet is reported at 2.1 GFLOPs, which is lower than the 2.3 GFLOPs of vanilla MobileNetV2, despite adding CBAM modules. This is surprising and likely a reporting artifact unless the same FLOPs counting method, input resolution, and module insertion points are specified. The efficiency claim ('only 2.1 GFLOPs', 'real-time deployment') is load-bearing for the paper's stated purpose. Please state exactly how FLOPs were computed, report values with and without CBAM for identical inputs, and, if 'real-time' is claimed, provide latency measurements on the target edge hardware.
[Tables I and II (statistical reliability)] All results are single-run point estimates with no standard deviations, confidence intervals, or repeated-seed experiments. The incremental gains are +2.2 points for CBAM and +1.1 points for two-phase training; with only 1,500 test images, these differences could easily be within run-to-run variance. The claim that both components 'significantly contribute' needs support. Please report means and standard deviations over at least three independent runs, and, if possible, a paired significance test or a per-class performance breakdown.
[Sections II.C and II.D (CBAM implementation)] The paper claims CBAM, which by definition includes both channel and spatial attention, but Equations (8)-(9) describe only a channel attention map M_c; no spatial attention module or its integration is defined. The text also refers to 'Hybrid Attention Injection' but does not specify the layer(s) in MobileNetV2 where the attention modules are inserted. Without these details, the architecture is not reproducible from the paper. Please provide a layer-level insertion diagram or pseudocode and include the spatial attention equations, or explicitly state that only channel attention is used.

minor comments (5)

[Equation (2)] The text says 'w_j is the class weight corresponding to the i-th sample'; it should be 'the class weight corresponding to class j.' The formula used to set the class weights (inverse frequency, etc.) should also be stated.
[Equation (10) vs Section III.A] Equation (10) describes an exponential learning-rate decay, while Section III.A fixes Phase 1 learning rate at 10^-3 and Phase 2 at 10^-4. The relationship between the scheduled rate and the two-phase rates needs clarification, including the number of epochs/iterations per phase.
[References] The original MobileNetV2 paper and the original ArcFace paper should be cited; the current [16] is an IoT application paper and [9] is a survey, not ArcFace.
[Table III baseline setup] For the 'independently implemented' baselines, the paper reports no hyperparameters (epochs, optimizer, batch size, data augmentation, learning-rate schedule). Saying 'identical training, validation, and testing splits' is not enough to guarantee a fair comparison.
[Section III.D] The Grad-CAM visualization is qualitative only; this is acceptable as an illustration, but the caption should state that no quantitative evaluation of attention maps is performed.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the reported accuracy is an empirical measurement, not a derivation from the paper's own definitions.

full rationale

ForensicNet's central claims are empirical measurements on a curated dataset, not quantities derived from the paper's equations. The mathematical statements in Sections II.A–II.E are standard definitions (dataset notation, weighted cross-entropy, freezing mask, MobileNetV2 features, CBAM channel attention, learning-rate schedule, classification head) and none of them is used to algebraically produce the accuracy, precision, recall, or F1 values in Tables I–III. The architecture comparison and ablation study are experimental, and the paper does not fit a parameter to a subset of data and then present a closely related quantity as a prediction. There are no load-bearing self-citations: the cited works are external references to standard components (MobileNetV2, CBAM, LFW, SCFace) and prior face-recognition methods, with no author-overlap uniqueness theorem or ansatz smuggled in by citation. The dataset description (15,000 images, 68 classes, 10,800 LFW + 4,200 SCFace) is internally hard to reconcile with SCFace's subject count, and the exact identity split is not released, but that is a reproducibility and correctness risk, not circularity. The paper itself acknowledges limitation to public benchmark datasets and controlled preprocessing. Because the headline result is a measured outcome rather than a derived equivalence, no circular step can be identified under the required standard of quoting a specific reduction.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central accuracy claim rests on no derivation; it is an empirical measurement, so what matters is the benchmark substrate and the hand-set training choices. The dataset curation (68 classes, 15,000 images, 80/10/10 split) is the largest unstated input, and the class weights of Eq. 2 plus the freezing threshold of Eq. 4 are never specified despite being load-bearing for the reported gains. No new entities are invented: ForensicNet is a named composition of published components, and 'adaptive layer unfreezing' is a relabeling of standard progressive unfreezing practice.

free parameters (4)

Class weights w_j in weighted cross-entropy (Eq. 2)
Never specified in Section II.A or III.A; material because the curated dataset mixes LFW (10,800 images) with SCFace (4,200 images) over a claimed 68 classes, implying strong per-class imbalance that the weights would need to counteract.
Freezing threshold L_0 (binary freeze mask, Eq. 4)
Defines the 'adaptive layer unfreezing' contribution credited with +1.1% (Table II), but the threshold value, the progressive unfreezing schedule, and epoch boundaries are not reported or ablated.
Two-phase learning rates and exponential decay schedule (Eq. 10) = 1e-3, then 1e-4 (Section III.A)
Hand-chosen constants (0.001, 0.96, floor(t/1000), and the /10 fine-tuning reduction) that define the two-phase training strategy whose contribution is the paper's main claim.
Data augmentation magnitudes = rotation +/-30 deg, occlusion mask 15-30%
Chosen by hand in Section II.A to simulate forensic degradation; not ablated, so the contribution of the augmentation recipe to the 92.4% result is unknown.

assumptions (4)

domain assumption ImageNet-pretrained MobileNetV2 features transfer to the forensic face domain
The entire transfer-learning strategy (Section II.B) presumes the pretrained backbone provides useful face features; it is never tested against training from scratch or a face-pretrained initialization.
domain assumption Closed-set accuracy on 68 curated identities measures forensic identification capability
The results (Section III) treat 68-class classification as forensic identification; the paper's own conclusion concedes that real-world generalization is untested.
domain assumption Selecting identity classes with at least 15 images yields a representative forensic benchmark
The filter in Section III.A selects the most-represented identities (top-frequency celebrities from LFW), which likely makes the closed-set task easier than unconstrained forensic surveillance; representativeness is asserted, not demonstrated.
standard math Standard CNN training and optimization framework (SGD/Adam, softmax, cross-entropy) applies as-is
Unproblematic background; Eqs. (2), (5), (10)-(12) presuppose the standard deep-learning optimization setup.

how reviews work

0 comments

Cite this review

Pith. "Pith review of ForensicNet: Lightweight Attention-Enhanced MobileNetV2 for Automated Face Identification." pith.science (2026). https://pith.science/paper/XTCXKG4X

@misc{pith2026260716273,
  author       = {Pith},
  title        = {Pith review of: ForensicNet: Lightweight Attention-Enhanced MobileNetV2 for Automated Face Identification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XTCXKG4X}},
  note         = {Machine review of arXiv:2607.16273}
}

read the original abstract

In forensic environments, automated identification of perpetrators is difficult due to pose changes, changes in light, occlusion, and lack of labeled data. This paper presents ForensicNet, a lightweight deep learning framework for forensic face recognition that enhances attention. The suggested model combines the MobileNetV2 backbone with Convolutional Block Attention Modules (CBAM) to improve the learning of discriminative features while maintaining computational speed. A two-phase transfer learning strategy with adaptive layer unfreezing is used to improve domain adaptation and reduce overfitting. This study used publicly available datasets such as LFW and SCFace, with 15,000 facial images spanning 68 identity classes. The proposed model outperforms baseline architectures such as AlexNet, ResNet-50, and MobileNetV2, with an accuracy of 92.4%, a precision of 90.8%, and a recall of 89.5%. Additionally, the framework requires only 2.1 GFLOPs per inference, and hence can be used in real-time forensic surveillance applications.

Figures

Figures reproduced from arXiv: 2607.16273 by the authors.

**Figure 1.** illustrates the overall architecture of the proposed ForensicNet, highlighting the integration of the MobileNetV2 backbone with CBAM attention modules and the final classification pipeline. A. Dataset and Preprocessing Let the forensic face dataset be represented as: , [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 4.** Training loss comparison. TABLE II. ABLATION STUDY Model Variant Accuracy (%) MobileNetV2 (Baseline) 89.1 + CBAM 91.3 +Two-Phase Training 92.4 C. Comparison with Recent Deep Learning Models To further evaluate the effectiveness of the proposed framework, additional comparisons were conducted with recent deep learning architectures commonly used in face recognition tasks, including EfficientNet, GhostNet, and Vision … view at source ↗

**Figure 2.** Accuracy comparison [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Figures from the paper (2 more)

**Figure 3.** Figure 3: Recall comparison [PITH_FULL_IMAGE:figures/full_fig_p004_3.png]

**Figure 5.** Figure 5: Anonymized Grad-CAM visualization highlighting discriminative attention regions on forensic facial images. IV. CONCLUSION This study presents ForensicNet, an efficient deep learning framework for automated forensic face identification under challenging surveillance con…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 9 canonical work pages

[1]

Study and design of rot ational and translational with low illumination based face imag e recognition,

B. G. Kumari and A. Gupta, "Study and design of rot ational and translational with low illumination based face imag e recognition," OPSEARCH , vol. 62, no. 4, pp. 1914–1937, Dec. 2025, https://doi.org/10.1007/s12597-024-00892-9

work page doi:10.1007/s12597-024-00892-9 1914
[2]

An Optimized Hybrid CNN- AlexNet Framework for Real-Time Crime Activity and Suspicious Object Detection,

R. Roopa and M. Humera Khanam, "An Optimized Hybrid CNN- AlexNet Framework for Real-Time Crime Activity and Suspicious Object Detection," in Proceedings of Sixth International Conference on Computer and Communication Technologies , vol. 1356, K. R. Madhavi, N. Ramrao, K. Kumar, K. S. Raju, and M. Sellathurai , Eds. Springer Nature Singapore, 2025, pp. 373–383

2025
[3]

A Comprehensive Exploration: A ttention Mechanisms in Facial Emotion Recognition,

Nidhi and B. Verma, "A Comprehensive Exploration: A ttention Mechanisms in Facial Emotion Recognition," in 2023 Seventh International Conference on Image Information Proce ssing (ICIIP) , Nov. 2023, pp. 591–596, https://doi.org/10.1109/ICIIP61524.2023.10537662

arXiv 2023
[4]

A Bro ad Study of Pre- training for Domain Generalization and Adaptation,

D. Kim, K. Wang, S. Sclaroff, and K. Saenko, "A Bro ad Study of Pre- training for Domain Generalization and Adaptation," in Computer Vision – ECCV 2022 , vol. 13693, S. Avidan, G. Brostow, M. Cissé, G. M . Farinella, and T. Hassner, Eds. Springer Nature Swi tzerland, 2022, pp. 621–638

2022
[5]

S phereFace Revived: Unifying Hyperspherical Face Recognition,

W. Liu, Y. Wen, B. Raj, R. Singh, and A. Weller, "S phereFace Revived: Unifying Hyperspherical Face Recognition," IEEE Transactions on Pattern Analysis and Machine Intelligence , vol. 45, no. 2, pp. 2458– 2474, Feb. 2023, https://doi.org/10.1109/TPAMI.2022 .3159732

doi:10.1109/tpami.2022 2023
[6]

A Comprehensive Review of Face Detection Techniques for Occluded Faces: Methods, Datasets, and Open Challen ges,

T. Thaher, M. Mafarja, M. Saffarini, A. H. H. M. Mo hamed, and A. A. El-Saleh, "A Comprehensive Review of Face Detection Techniques for Occluded Faces: Methods, Datasets, and Open Challen ges," Computer Modeling in Engineering & Sciences , vol. 143, no. 3, pp. 2615–2673, 2025, https://doi.org/10.32604/cmes.2025.064857

arXiv 2025
[7]

Sohail, S

S. Sohail, S. M. Sajjad, A. Zafar, Z. Iqbal, Z. Muh ammad, and M. Kazim, "Deepfake Image Forensics for Privacy Protec tion and Engineering, Technology & Applied Science Research Vol. 16 , No. 4, 20 26 , 37620 -37625 37625 www.etasr.com N. J. & B. T : ForensicNet: Lightweight Attention -Enhanced MobileNetV2 for Automated Face … Authenticity Using Deep Lear...

work page doi:10.3390/info160 2025
[8]

Impact Analysis of Different Effective Loss Functi ons by Using Deep Convolutional Neural Network for Face Recognition,

A. D. Nguyen, D. T. Nguyen, H. N. Dao, H. H. Le, an d N. Q. Tran, "Impact Analysis of Different Effective Loss Functi ons by Using Deep Convolutional Neural Network for Face Recognition," in From Born- Physical to Born-Virtual: Augmenting Intelligence i n Digital Libraries , vol. 13636, Y. H. Tseng, M. Katsurai, and H. N. Ngu yen, Eds. Springer Internatio...

2022

Show all 23 references

[9]

A Comparative Review and Performance Benchmarking of Face Recognition Models,

A. K. Mohapatra, K. Kumar, S. Swarup, H. Sharma, an d V. Sinha, "A Comparative Review and Performance Benchmarking of Face Recognition Models," in Proceedings of Data Analytics and Management , vol. 1602, A. Swaroop, B. Virdee, S. Correia, and Z. Polkowski, Eds. Springer Natur...

2026
[10]

Conv-MTD: A CNN Based Multi-Label Medic al Tubes Detection and Classification Model to Facilitate Re source-Constrained Point-of-Care Devices,

M. Abbas, W. C. Kuo, K. Mahmood, W. Akram, S. Mehmo od, and A. K. Bashir, "Conv-MTD: A CNN Based Multi-Label Medic al Tubes Detection and Classification Model to Facilitate Re source-Constrained Point-of-Care Devices," IEEE Journal of Biomedical and Health Informatics , vol. 3...

2026
[11]

D eep Learning- Based Brain Tumour Detection: A Comparative Study o f ResNet50, AlexNet, and SqueezeNet,

M. Rajput, S. Ponde, S. Tamane, and S. Deshmukh, "D eep Learning- Based Brain Tumour Detection: A Comparative Study o f ResNet50, AlexNet, and SqueezeNet," in 2025 International Conference on Applications of Machine Intelligence and Data Analy tics (ICAMIDA) , Aug. 2025, pp. 1...

2025
[12]

A lightwe ight CNN-based algorithm and implementation on embedded system for real-time face recognition,

Z. Chen, J. Chen, G. Ding, and H. Huang, "A lightwe ight CNN-based algorithm and implementation on embedded system for real-time face recognition," Multimedia Systems , vol. 29, no. 1, pp. 129–138, Feb. 2023, https://doi.org/10.1007/s00530-022-00973-z

2023 doi
[13]

FD REnet: Face Detection and Recognition Pipeline,

D. Virmani, P. Girdhar, P. Jain, and P. Bamdev, "FD REnet: Face Detection and Recognition Pipeline," Engineering, Technology & Applied Science Research , vol. 9, no. 2, pp. 3933–3938, Apr. 2019, https://doi.org/10.48084/etasr.2492

2019 doi
[14]

A survey on face da ta augmentation for the training of deep neural networks,

X. Wang, K. Wang, and S. Lian, "A survey on face da ta augmentation for the training of deep neural networks," Neural Computing and Applications , vol. 32, no. 19, pp. 15503–15531, Oct. 2020, https://doi.org/10.1007/s00521-020-04748-3

2020 doi
[15]

Hybri d CNN-BiLSTM Architecture with Transfer Learning and Fine-Tuning for Efficient Deepfake Video Detection on Celeb-Df v2,

P. Bakshi, S. Verma, B. Rawat, and E. Verma, "Hybri d CNN-BiLSTM Architecture with Transfer Learning and Fine-Tuning for Efficient Deepfake Video Detection on Celeb-Df v2," in 2025 IEEE International Conference on Advances in Computing Research On Sci ence Engineering and Tech...

2025
[16]

IoT-MFaceNet: Internet-of-Thin gs-Based Face Recognition Using MobileNetV2 and FaceNet Deep-Lear ning Implementations on a Raspberry Pi-400,

A. S. Mohammad, T. G. Jarullah, M. T. S. Al-Kaltakc hi, J. Alshehabi Al-Ani, and S. Dey, "IoT-MFaceNet: Internet-of-Thin gs-Based Face Recognition Using MobileNetV2 and FaceNet Deep-Lear ning Implementations on a Raspberry Pi-400," Journal of Low Power Electronics and Applicat...

2024 doi
[17]

AI-powered Forensic Science - E videnceOptimizer: Leveraging Hybrid Ensembled Models,

U. Ragavee, C. Harini, K. G. Dhanushri, Y. R. Jaisr i, D. Nitesh Singh, and M. V. Shankar, "AI-powered Forensic Science - E videnceOptimizer: Leveraging Hybrid Ensembled Models," in 2025 International Conference on Smart & Sustainable Technology (INCSS T) , July 2025, pp. 1–7,...

2025
[18]

Yolov7-pcbam: enhancing steel surface defect detection via partial convolution and attent ion mechanism,

Y. H. Shih and J. H. Chen, "Yolov7-pcbam: enhancing steel surface defect detection via partial convolution and attent ion mechanism," Machine Vision and Applications , vol. 37, no. 2, Mar. 2026, Art. no. 29, https://doi.org/10.1007/s00138-025-01784-8

2026 doi
[19]

Advances in Face Recognition: A Comprehensive Review of Feat ure Extraction and Dataset Evaluation,

S. M. H. Abidi, S. A. Hassan, S. M. Raza, and M. J. Beliatis, "Advances in Face Recognition: A Comprehensive Review of Feat ure Extraction and Dataset Evaluation," Electronics , vol. 15, no. 2, Jan. 2026, Art. no. 338, https://doi.org/10.3390/electronics15020338

2026 doi
[20]

Learning to Align from Scratch,

G. Huang, M. Mattar, H. Lee, and E. G. Learned-mill er, "Learning to Align from Scratch," in Advances in Neural Information Processing Systems , 2012, vol. 25

2012
[21]

LFW Dataset – Machine Learning Datasets

"LFW Dataset – Machine Learning Datasets." [Online] . Available: https://datasets-dev.10web.me/docs/ml/datasets/lfw- dataset/
[22]

SCface – surveil lance cameras face database,

M. Grgic, K. Delac, and S. Grgic, "SCface – surveil lance cameras face database," Multimedia Tools and Applications , vol. 51, no. 3, pp. 863– 879, Feb. 2011, https://doi.org/10.1007/s11042-009- 0417-2

2011 doi
[23]

SCface - Surveillance Cameras Face Database

"SCface - Surveillance Cameras Face Database." [Onl ine]. Available: https://www.scface.org/

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.

[1] [1]

Study and design of rot ational and translational with low illumination based face imag e recognition,

B. G. Kumari and A. Gupta, "Study and design of rot ational and translational with low illumination based face imag e recognition," OPSEARCH , vol. 62, no. 4, pp. 1914–1937, Dec. 2025, https://doi.org/10.1007/s12597-024-00892-9

work page doi:10.1007/s12597-024-00892-9 1914

[2] [2]

An Optimized Hybrid CNN- AlexNet Framework for Real-Time Crime Activity and Suspicious Object Detection,

R. Roopa and M. Humera Khanam, "An Optimized Hybrid CNN- AlexNet Framework for Real-Time Crime Activity and Suspicious Object Detection," in Proceedings of Sixth International Conference on Computer and Communication Technologies , vol. 1356, K. R. Madhavi, N. Ramrao, K. Kumar, K. S. Raju, and M. Sellathurai , Eds. Springer Nature Singapore, 2025, pp. 373–383

2025

[3] [3]

A Comprehensive Exploration: A ttention Mechanisms in Facial Emotion Recognition,

Nidhi and B. Verma, "A Comprehensive Exploration: A ttention Mechanisms in Facial Emotion Recognition," in 2023 Seventh International Conference on Image Information Proce ssing (ICIIP) , Nov. 2023, pp. 591–596, https://doi.org/10.1109/ICIIP61524.2023.10537662

arXiv 2023

[4] [4]

A Bro ad Study of Pre- training for Domain Generalization and Adaptation,

D. Kim, K. Wang, S. Sclaroff, and K. Saenko, "A Bro ad Study of Pre- training for Domain Generalization and Adaptation," in Computer Vision – ECCV 2022 , vol. 13693, S. Avidan, G. Brostow, M. Cissé, G. M . Farinella, and T. Hassner, Eds. Springer Nature Swi tzerland, 2022, pp. 621–638

2022

[5] [5]

S phereFace Revived: Unifying Hyperspherical Face Recognition,

W. Liu, Y. Wen, B. Raj, R. Singh, and A. Weller, "S phereFace Revived: Unifying Hyperspherical Face Recognition," IEEE Transactions on Pattern Analysis and Machine Intelligence , vol. 45, no. 2, pp. 2458– 2474, Feb. 2023, https://doi.org/10.1109/TPAMI.2022 .3159732

doi:10.1109/tpami.2022 2023

[6] [6]

A Comprehensive Review of Face Detection Techniques for Occluded Faces: Methods, Datasets, and Open Challen ges,

T. Thaher, M. Mafarja, M. Saffarini, A. H. H. M. Mo hamed, and A. A. El-Saleh, "A Comprehensive Review of Face Detection Techniques for Occluded Faces: Methods, Datasets, and Open Challen ges," Computer Modeling in Engineering & Sciences , vol. 143, no. 3, pp. 2615–2673, 2025, https://doi.org/10.32604/cmes.2025.064857

arXiv 2025

[7] [7]

Sohail, S

S. Sohail, S. M. Sajjad, A. Zafar, Z. Iqbal, Z. Muh ammad, and M. Kazim, "Deepfake Image Forensics for Privacy Protec tion and Engineering, Technology & Applied Science Research Vol. 16 , No. 4, 20 26 , 37620 -37625 37625 www.etasr.com N. J. & B. T : ForensicNet: Lightweight Attention -Enhanced MobileNetV2 for Automated Face … Authenticity Using Deep Lear...

work page doi:10.3390/info160 2025

[8] [8]

Impact Analysis of Different Effective Loss Functi ons by Using Deep Convolutional Neural Network for Face Recognition,

A. D. Nguyen, D. T. Nguyen, H. N. Dao, H. H. Le, an d N. Q. Tran, "Impact Analysis of Different Effective Loss Functi ons by Using Deep Convolutional Neural Network for Face Recognition," in From Born- Physical to Born-Virtual: Augmenting Intelligence i n Digital Libraries , vol. 13636, Y. H. Tseng, M. Katsurai, and H. N. Ngu yen, Eds. Springer Internatio...

2022