Spatial Attention Supervision for Defect Localization: Exploiting Ground-Truth Masks as Training Signal in Diffusion-Augmented Defect Detection

Sajjad Rezvani Boroujeni; Muskan Saraf; Gnana Tulasi Makineni; Tom Bush; Hossein Abedi

cs.CVSep 5, 2026

Spatial Attention Supervision for Defect Localization: Exploiting Ground-Truth Masks as Training Signal in Diffusion-Augmented Defect Detection

Authors: Sajjad Rezvani Boroujeni, Muskan Saraf, Gnana Tulasi Makineni, Tom Bush, Hossein Abedi

Organizations: Data Science Department, Actual Reality Technologies, OH, USA.

Abstract

Ground-truth defect masks in industrial inspection datasets are typically reserved for evaluation. This paper repurposes them as spatial supervision signals during training of classification networks, teaching a model not just what to predict but where to look. The method adds an activation-based attention alignment loss that steers convolutional feature maps toward defect regions, in a mixed-supervision formulation that also accommodates samples without masks, such as diffusion-generated images. Combined with DDPM augmentation, synthetic images contribute quantity while masks contribute spatial precision. We evaluate 85 models (four CNN backbones under a 2x2 data/training factorial over five seeds, plus a Swin-V2-T transformer baseline) on the MVTec-AD bottle benchmark, with localization measured on held-out defect images excluded from classifier gradient updates. Main findings: (1) attention-guided training improves activation-based localization (Pixel-AUROC) by +18.0% for EfficientNetB0 with augmentation (p=0.005, Cohen's d=2.6) and +18.7% for ResNet50 (p=0.008), significant in four of eight CNN settings (uncorrected for multiple comparisons) with no significant change in classification; (2) for EfficientNetB0 a data x training-mode interaction is significant (p=0.002), consistent with a super-additive effect (+13.6% combined vs +1.6% summed individual effects); (3) architectures with weaker spatial representations benefit most, whereas ConvNeXt-T shows no effect, apparently because its depthwise-convolution activations yield spatially uninformative channel-mean maps; (4) unsupervised PatchCore remains the strongest localizer (Pixel-AUROC=0.983), contextualizing the supervised gains. These results show that existing evaluation masks can act as practical training signals that measurably and reproducibly improve where defect classifiers attend.

Figures & tables

Approach	Use of Masks	Difference from Proposed
Segmentation (U-Net, DeepLab)	Direct output target	Predicts mask; typically trained with dense mask supervision
Ross et al. [ 15 ]	Gradient penalty	Input gradients (not activations); not industrial
Kasem et al. [ 13 ]	None (architectural)	CBAM modules in architecture, not a training loss
Region-Aware CAM [ 14 ]	Evaluation only	Weakly-supervised; no mask-guided training
PatchCore [ 20 ]	Evaluation only	Unsupervised; masks unused during training
Prior work [ 3 ]	Evaluation only	DDPM augmentation for classification; no localization

Table 1: Positioning relative to prior work. Among the approaches reviewed, the proposed method is distinctive in combining auxiliary mask supervision with diffusion augmentation in a mixed-supervision setting.

Figure 1: Pipeline overview. Data: MVTec-AD bottle with optional DDPM augmentation. Training: 85 models total, four CNN backbones under a $2\times 2$ factorial (data setting $\times$ training mode) across five seeds, plus Swin-V2-T baseline across five seeds. Evaluation: held-out classification metrics and held-out localization via Pixel-AUROC.

Backbone	Type	Year	Params	Input Size
EfficientNetB0 [ 25 ]	CNN	2019	5.3M	224 $\times$ 224
MobileNetV2 [ 26 ]	CNN	2018	3.5M	224 $\times$ 224
ResNet50 [ 27 ]	CNN	2016	25.6M	224 $\times$ 224
ConvNeXt-T [ 28 ]	CNN	2022	28.6M	224 $\times$ 224
Swin-V2-T [ 29 ]	Transformer	2022	28M	256 $\times$ 256

Table 2: Backbone Architectures.

Backbone	Config	F1	AUROC	Recall
EfficientNetB0	Real, Std	.888 $\pm$ .072	.961 $\pm$ .048	.911
	Real, AG	.878 $\pm$ .086	.949 $\pm$ .066	.911
	Aug, Std	.902 $\pm$ .063	.975 $\pm$ .016	.842
	Aug, AG	.886 $\pm$ .052	.977 $\pm$ .012	.827
MobileNetV2	Real, Std	.844 $\pm$ .067	.962 $\pm$ .031	.867
	Real, AG	.878 $\pm$ .058	.971 $\pm$ .018	.889

Table 3: Held-out classification performance (F1 and AUROC as mean

\pm

std across 5 seeds; Recall as mean). Best F1 per backbone in bold. For ConvNeXt-T, standard and attention-guided training yield classification metrics that coincide to three decimal places (the alignment loss produces no discernible change in its classification; see Section 6.1 ), so the two modes are reported jointly as “Std/AG.”

Figure 2: ( a ) Held-out classification F1 across four training configurations for three CNN architectures. Swin-V2-T reference line at F1=1.000. ( b ) Best configuration per architecture, showing ConvNeXt-T and Swin-V2-T dominate classification while lighter CNNs achieve 0.88 to 0.90 F1.

Backbone	Data	Std	AG	$\Delta$	$p$	$d$
EfficientNetB0	Real	0.552 $\pm$ .047	0.582 $\pm$ .058	+5.4%	0.051	1.23
EfficientNetB0	+DDPM	0.531 $\pm$ .063	0.627 $\pm$ .040	+18.0%	0.005	2.57
MobileNetV2	Real	0.563 $\pm$ .053	0.585 $\pm$ .055	+4.0%	0.020	1.68
MobileNetV2	+DDPM	0.562 $\pm$ .054	0.588 $\pm$ .044	+4.5%	0.238	0.62
ResNet50	Real	0.464 $\pm$ .045	0.550 $\pm$ .049	+18.7%	0.008	2.18
ResNet50	+DDPM	0.454 $\pm$ .040	0.509 $\pm$ .013	+12.0%	0.023	1.61

Table 4: Held-out localization (activation-based Pixel-AUROC, mean

\pm

std across 5 seeds). Each model is evaluated only on held-out validation defect images excluded from its gradient updates.

\Delta

is the relative change from standard (Std) to attention-guided (AG);

p

is a paired

t

-test (by seed) and

d

is Cohen’s

d

. Significant improvements (

p<0.05

, uncorrected for multiple comparisons) in bold. Swin-V2-T (occlusion) and PatchCore (unsupervised) are localization references. 0 0 footnotetext: † ConvNeXt-T shows no consistent effect and very high seed variance; a plausible explanation is that its depthwise-convolution activations are spatially uninformative, which would leave the alignment loss with little useful gradient (Section 6.1 ).

Figure 3: ( a ) Held-out activation-based localization (Pixel-AUROC): standard vs. attention-guided across three CNN architectures and two data settings, with error bars (std across 5 seeds) and the PatchCore reference line. ( b ) $2\times 2$ decomposition for EfficientNetB0: DDPM augmentation and attention guidance together show a significant super-additive interaction ( $p=0.002$ ; Section 5.3 ).

Figure 4: Held-out activation-based vs. Grad-CAM localization for the four CNN backbones (AugmentedData-trained, evaluated on held-out defect images). Arrows show the shift from standard (circle) to attention-guided (square) training. All points fall below the diagonal, indicating that Grad-CAM outperforms raw activation maps. Arrows point upward for EfficientNetB0, MobileNetV2, and ResNet50, showing that attention guidance improves activation-based localization while leaving Grad-CAM largely unchanged. ConvNeXt-T’s arrow does not point up: its activation map is spatially uninformative and high-variance, and the alignment loss does not improve it, even though its Grad-CAM localization is the strongest of all backbones (see Section 6.1 ).

Backbone	Data	Std	AG	$\Delta$	$p$	$d$
EfficientNetB0	Real	0.749 $\pm$ .041	0.753 $\pm$ .041	+0.5%	0.526	0.31
EfficientNetB0	+DDPM	0.752 $\pm$ .064	0.772 $\pm$ .039	+2.6%	0.323	0.50
MobileNetV2	Real	0.623 $\pm$ .086	0.623 $\pm$ .063	0.0%	0.993	0.00
MobileNetV2	+DDPM	0.696 $\pm$ .073	0.703 $\pm$ .051	+1.1%	0.714	0.18
ResNet50	Real	0.543 $\pm$ .023	0.579 $\pm$ .025	+6.6%	0.044	1.30
ResNet50	+DDPM	0.485 $\pm$ .044	0.537 $\pm$ .026	+10.7%	0.024	1.59

Table 5: Held-out Grad-CAM localization (Pixel-AUROC, mean

\pm

std across 5 seeds), reported as in Table 4 .

\Delta

is the standard

\rightarrow

attention-guided change;

p

is a paired

t

-test (by seed) and

d

is Cohen’s

d

. Grad-CAM values exceed the activation-based ones (cf. Table 4 ); attention guidance transfers to Grad-CAM significantly only for ResNet50 (significant improvements in bold).

Figure 5: Qualitative comparison of activation heatmaps for EfficientNetB0 (AugmentedData). Columns: input image, ground-truth mask, standard training heatmap, attention-guided heatmap. Attention-guided models produce activations that are visually more concentrated on the actual defect regions.

Industrial Anomaly Detection

Industrial anomaly detection finds defective products and abnormal equipment behavior from visual, sensor, audio, and other production data.

See all 261 papers

Industrial Surface Defect Detection via Diffusion Generation and Asymmetric Student-Teacher Network
Apr 21, 2026Shuo Feng, Runlin Zhou, Yuyang Li +1Industrial Anomaly Detection Synthetic Anomaly Generation
School of Automation, Southeast University, Nanjing, China
OSAGEN: Object-Aware Mask Priors and Multistage Decoupled Diffusion for Industrial Anomaly Generation
Jul 31, 2026Jinyi Xu, Peng Chen, Yunkang Cao +3Industrial Anomaly Detection Anomaly Localization
School of Cyber Science and Technology, Shenzhen Campus of Sun Yat-sen University · School of Artificial Intelligence and Robotics, Hunan University · Department of Computer and Information Science, University of Macau +1
Boxes2Pixels: Learning Defect Segmentation from Noisy SAM Masks
Apr 13, 2026Camile Lendering, Erkut Akdag, Egor BondarevWeakly Supervised Image Segmentation Semantic Segmentation
AIMS Group, Department of Electrical Engineering, Eindhoven University of Technology

Apr 21, 2026J/K move · Enter open · S save

Industrial Surface Defect Detection via Diffusion Generation and Asymmetric Student-Teacher Network

Shuo Feng, Runlin Zhou, Yuyang Li, Guangcan Liu

School of Automation, Southeast University, Nanjing, China

Industrial Anomaly Detection Synthetic Anomaly Generation Teacher-Student Learning

Open paper AskPDF arXiv

Industrial surface defect detection often suffers from limited defect samples, severe long-tailed distributions, and difficulties in accurately localizing subtle defects under complex backgrounds. To address these challenges, this paper proposes an unsupervised defect detection method that integrates a Denoising Diffusion Probabilistic Model (DDPM) with an asymmetric teacher-student architecture. First, at the data level, the DDPM is trained solely on normal samples. By introducing constant-variance Gaussian perturbations and Perlin noise-based masks, high-fidelity and physically consistent defect samples along with pixel-level annotations are generated, effectively alleviating the data scarcity problem. Second, at the model level, an asymmetric dual-stream network is constructed. The teacher network provides stable representations of normal features, while the student network reconstructs normal patterns and amplifies discrepancies between normal and anomalous regions. Finally, a joint optimization strategy combining cosine similarity loss and pixel-wise segmentation supervision is adopted to achieve precise localization of subtle defects. Experimental results on the MVTecAD dataset show that the proposed method achieves 98.4% image-level AUROC and 98.3% pixel-level AUROC, significantly outperforming existing unsupervised and mainstream deep learning methods. The proposed approach does not require large amounts of real defect samples and enables accurate and robust industrial defect detection and localization. \keywords{Industrial defect detection \and diffusion models \and data generation \and teacher-student architecture \and pixel-level localization}

Abstract

Industrial Anomaly Detection

Explore similar work

Industrial Surface Defect Detection via Diffusion Generation and Asymmetric Student-Teacher Network

OSAGEN: Object-Aware Mask Priors and Multistage Decoupled Diffusion for Industrial Anomaly Generation

Boxes2Pixels: Learning Defect Segmentation from Noisy SAM Masks