PaSTe: Improving the Efficiency of Visual Anomaly Detection at the Edge

This repository contains supplementary plots and tables related to the work presented in the paper titled: "PaSTe: Improving the Efficiency of Visual Anomaly Detection at the Edge".

Code repository

This project gave life to an open-source library for Visual Anomaly Detection, which is modular in nature: MoViAD. That library contains the implementation of the algorithms used in this work, as well as the code for training and evaluating them.

git clone https://github.com/AMCO-UniPD/moviad

Abstract

Visual Anomaly Detection (VAD) has gained significant research attention for its ability to identify anomalous images and pinpoint the specific areas responsible for the anomaly. A key advantage of VAD is its unsupervised nature, which eliminates the need for costly and time-consuming labeled data collection. However, despite its potential for real-world applications, the literature has given limited focus to resource-efficient VAD, particularly for deployment on edge devices. This work addresses this gap by leveraging lightweight neural networks to reduce memory and computation requirements, enabling VAD deployment on resource-constrained edge devices. We benchmark the major VAD algorithms within this framework and demonstrate the feasibility of edge-based VAD using the well-known MVTec dataset. Furthermore, we introduce a novel algorithm, Partially Shared Teacher-student PaSTe, designed to address the high resource demands of the existing Student Teacher Feature Pyramid Matching (STFPM) approach. Our results show that PaSTe decreases the inference time by 25%, while reducing the training time by 33% and peak RAM usage during training by 76%.

Citation

@inproceedings{,
    title={},
    author={},
    booktitle={},
    pages={},
    year={},
    organization={}
}

Supplementary material

Experimental setting

Backbones details

• MobileNet V2: Available in PyTorch, using the ImageNet-pretrained version from TorchVision.
• MCUNet: PyTorch-based implementation of the MCUNet-in3 network, pretrained on ImageNet and available on the official GitHub repository.
• PhiNet: PyTorch-based implementation trained on ImageNet, with source code available in the MicroMind repository. The considered PhiNet uses the following hyperparameters: num_layers = 7, alpha = 1.2, beta = 0.5, t0 = 6.
• MicroNet: PyTorch-based implementation of the MicroNet-m1 network, pretrained on ImageNet. Network weights are available on the official GitHub repository, while the architecture code has been refactored by us.

Computational resources

The characteristics of the machine on which we run the experiments are as follows:

Component	Specification
CPU	AMD Ryzen Threadripper 1920X 12- (24) @ 3.500GHz
RAM	125GiB
GPU	NVIDIA TITAN V
OS	Ubuntu 18.04.6 LTS x86_64

Feature extraction layers grid search

For MVTec dataset, we performed grid search on different groups of layers, to test the feature extraction at different depths. Th following table shows the grop of layers used for feature extraction, for each backbone in the grid search. Low, Mid, High refer to the depth of the layer group in the particular backbone architecture. Equiv refers to the layers which are equivalent in terms of %MACs to the first three of the WideResnet 50 backbone. Finally, PaSTe refers to the same layers as in the Equiv group, but the first layers have been shifted to account for the Partial Teacher Sharing technique. Last refers to the index of the last layer of the feature extraction backbone

	PhiNet	MicroNet	MCUNet	MobileNet V2
Low	[4, 5, 6]	[1, 2, 3]	[3, 6, 9]	[4, 7, 10]
Mid	[5, 6, 7]	[2, 3, 4]	[6, 9, 12]	[7, 10, 13]
High	[6, 7, 8]	[3, 4, 5]	[9, 12, 15]	[10, 13, 16]
Equiv	[2, 6, 7]	[2, 4, 5]	[2, 6, 14]	[3, 8, 14]
PaSTe	[5, 6, 7]	[3, 4, 5]	[6, 10, 14]	[7, 10, 14]
Last	9	7	17	18

Table 1

This table reports for every AD model and for every backbone the set of layers that achieve the highest F1 pixel level score.

AD Model	Backbone	Feature Layers	F1 pixel
CFA	MCUNet	[3, 6, 9]	0.572208
CFA	MicroNet	[1, 2, 3]	0.218769
CFA	MobileNet	[3, 8, 14]	0.553594
CFA	PhiNet	[4, 5, 6]	0.539773
PaDiM	MCUNet	[6, 9, 12]	0.502408
PaDiM	MicroNet	[1, 2, 3]	0.234883
PaDiM	MobileNet	[4, 7, 10]	0.533622
PaDiM	PhiNet	[4, 5, 6]	0.46044
PatchCore	MCUNet	[3, 6, 9]	0.530831
PatchCore	MicroNet	[2, 3, 4]	0.166413
PatchCore	MobileNet	[3, 8, 14]	0.526729
PatchCore	PhiNet	[4, 5, 6]	0.552885
STFPM	MCUNet	[6, 9, 12]	0.553616
STFPM	MicroNet	[2, 3, 4]	0.176685
STFPM	MobileNet	[4, 7, 10]	0.552699
STFPM	PhiNet	[5, 6, 7]	0.473028

If you are interested in the F1 pixel level and other metrics scores obtained by the different AD models with different bakcbones and different layers on the different MVTec categories, you can see this Table

Table 3 Extension

This is an extension of the Table 3 present in the paper. In this table all the backbones are reported.

		PatchCore	PaDiM	CFA	STFPM	PaSTe
	Total Memory [MB]	300	3.72G	141	189.7	-
WideResNet50	Inference [MAC]	10.42G	9.19G	36.89G	18.3G	-
	AD Performance [F1]	0.57	0.57	0.60	0.51	-
	Total Memory [MB]	31.11	49.4	6.16	5.32	5.11
MobileNetV2	Inference [MAC]	235.6M	225.5M	2.8G	454.4M	341.2M
	AD Performance [F1]	0.53	0.49	0.55	0.52	0.53
	Total Memory [MB]	44.37	428.7	20.45	2.7	2.5
PhiNet	Inference [MAC]	2.17G	217M	3.2G	433.8M	271.7M
	AD Performance [F1]	0.44	0.43	0.48	0.47	0.47
	Total Memory [MB]	13.1	2.73	0.727	0.6	0.59
MicroNet-m1	Inference [MAC]	29.3M	2.93M	817M	5.86M	0.58M
	AD Performance [F1]	0.16	0.2	0.07	0.17	0.17
	Total Memory [MB]	16.48	15.12	2.69	1.76	1.68
MCUNet-in3	Inference [MAC]	1.1G	112M	2.9G	224.3M	156.9M
	AD Performance [F1]	0.53	0.45	0.55	0.52	0.52

Overall benchmark complete plot

This plot also appears in the paper without the Micronet backbone, but what follows is the complete verison.

Plot 1

This serie of line plots, given an AD model and a backbone, hihglight the F1 pixel level score of different layers on different categories.

Plot 2

This serie of line plots, given an AD model, hihglight the F1 pixel level score of the best layer for every backbone on different categories.

PaSTe Performances VS Anomaly Size

The anomalies in the images have different sizes depending on the category in the MVTec AD Dataset.

By computing the mean area of anomalous pixels thanks to the ground truth present in the dataset, we can estimate the "mean anomaly size" of each category. The following plot shows the mean anomaly area for each category, in decreasing order.

The following image helps to visualize the performances of PaSTe for each category and each backbone, compared to the original STFPM method, showing the small overall performance degradation of the method.

All Evaluation Metrics - Equivalent Layers

We report all the metrics which are the most common in the context of VAD, to compare with other methods. The metrics are referred either as Image or Pixel, depending on whether the labels are the singular pixels or the whole images.

Method	Backbone	Layers	AUROC Image	AUROC Pixel	F1 Image	F1 Pixel	PRAUC Img	PRAUC Pixel	AUPRO Pixel
PaDiM	MCUNet-in3	[2, 6, 14]	0.90	0.95	0.93	0.44	0.95	0.37	0.85
PaDiM	MicroNet-m1	[2, 4, 5]	0.65	0.75	0.84	0.19	0.82	0.14	0.64
PaDiM	MobileNetV2	[3, 8, 14]	0.90	0.96	0.93	0.48	0.95	0.41	0.89
PaDiM	PhiNet	[2, 6, 7]	0.90	0.95	0.93	0.42	0.94	0.34	0.84
PaDiM	WideResNet50	[1, 2, 3]	0.95	0.97	0.95	0.57	0.97	0.53	0.93
STFPM	MCUNet-in3	[2, 6, 14]	0.89	0.95	0.93	0.52	0.95	0.51	0.84
STFPM	MicroNet-m1	[2, 4, 5]	0.61	0.73	0.84	0.16	0.81	0.10	0.57
STFPM	MobileNetV2	[3, 8, 14]	0.88	0.95	0.93	0.51	0.95	0.51	0.82
STFPM	PhiNet	[2, 6, 7]	0.86	0.94	0.91	0.47	0.93	0.46	0.77
STFPM	WideResNet50	[1, 2, 3]	0.85	0.95	0.92	0.50	0.92	0.45	0.89
CFA	MCUNet-in3	[2, 6, 14]	0.89	0.96	0.93	0.55	0.96	0.53	0.89
CFA	MicroNet-m1	[2, 4, 5]	0.56	0.51	0.84	0.06	0.77	0.03	0.50
CFA	MobileNetV2	[3, 8, 14]	0.93	0.97	0.95	0.55	0.97	0.53	0.89
CFA	PhiNet.5_6	[2, 6, 7]	0.90	0.95	0.93	0.47	0.96	0.45	0.84
CFA	WideResNet50	[1, 2, 3]	0.98	0.98	0.98	0.60	0.99	0.59	0.93
PatchCore	MCUNet-in3	[2, 6, 14]	0.97	0.95	0.96	0.52	0.98	0.51	0.88
PatchCore	MicroNet-m1	[2, 4, 5]	0.69	0.71	0.85	0.16	0.85	0.10	0.67
PatchCore	MobileNetV2	[3, 8, 14]	0.97	0.95	0.96	0.52	0.99	0.52	0.86
PatchCore	PhiNet5_6	[2, 6, 7]	0.94	0.93	0.94	0.43	0.97	0.40	0.82
PatchCore	WideResNet50	[1, 2, 3]	0.98	0.97	0.97	0.57	0.99	0.57	0.91

ViSA Benchmark

We provide the results of the experiments conducted on the ViSA Dataset. We consider the equivalent AD layers group.

ad_model	backbone	category	F1 pixel	F1 image	Image ROC AUC	Per-pixel ROC AUC	PR AUC image	PR AUC pixel	AU PRO pixel
CFA	MCUNet	Candle	0.231305	0.848571	0.88875	0.974044	0.904507	0.120537	0.911716
CFA	MCUNet	Capsules	0.159513	0.785858	0.570167	0.900671	0.69492	0.0611	0.744196
CFA	MCUNet	Cashew	0.741117	0.853068	0.8417	0.996184	0.919124	0.763404	0.862632
CFA	MCUNet	Chewinggum	0.258129	0.948362	0.9748	0.989939	0.988937	0.154468	0.911783
CFA	MCUNet	Fryum	0.478372	0.858253	0.8325	0.96699	0.917923	0.429479	0.92766
CFA	MCUNet	Macaroni1	0.0176119	0.741826	0.75415	0.987492	0.7474	0.00609797	0.933025
CFA	MCUNet	Macaroni2	0.00833604	0.693188	0.62945	0.853579	0.610905	0.000500356	0.779457
CFA	MCUNet	Pcb1	0.675444	0.862383	0.9068	0.988685	0.909859	0.70624	0.954699
CFA	MCUNet	Pcb2	0.173599	0.829589	0.8937	0.981571	0.910147	0.0986074	0.932107
CFA	MCUNet	Pcb3	0.358251	0.855124	0.899059	0.992115	0.917147	0.266452	0.926475
CFA	MCUNet	Pcb4	0.384871	0.954659	0.987426	0.975652	0.985725	0.387022	0.89747
CFA	MCUNet	Pipe_fryum	0.656682	0.953751	0.9704	0.99544	0.984635	0.673298	0.950646
CFA	MicroNet-M1	Candle	0.00701108	0.713922	0.617075	0.753208	0.632116	0.00220577	0.570341
CFA	MicroNet-M1	Capsules	0.0120705	0.772212	0.440625	0.661585	0.624147	0.0042895	0.527559
CFA	MicroNet-M1	Cashew	0.0233718	0.801606	0.2826	0.538554	0.59898	0.0114953	0.384105
CFA	MicroNet-M1	Chewinggum	0.00583875	0.810596	0.6087	0.534727	0.769513	0.00214929	0.507332
CFA	MicroNet-M1	Fryum	0.0634203	0.831643	0.68125	0.544186	0.811181	0.0274248	0.393191
CFA	MicroNet-M1	Macaroni1	0.000218058	0.738904	0.750925	0.439374	0.745004	8.00439e-05	0.49654
CFA	MicroNet-M1	Macaroni2	0.000176037	0.692608	0.58235	0.475757	0.562619	6.67328e-05	0.474811
CFA	MicroNet-M1	Pcb1	0.0102466	0.702535	0.628025	0.590941	0.657868	0.00423738	0.332694
CFA	MicroNet-M1	Pcb2	0.00336491	0.758943	0.828225	0.547471	0.871989	0.0012199	0.54362
CFA	MicroNet-M1	Pcb3	0.00358715	0.699245	0.693366	0.505946	0.708611	0.00150192	0.333585
CFA	MicroNet-M1	Pcb4	0.0104989	0.713126	0.592129	0.420318	0.567244	0.00376117	0.43914
CFA	MicroNet-M1	Pipe_fryum	0.0220462	0.846075	0.80955	0.425121	0.891249	0.00941748	0.436977
CFA	MobileNetV2	Candle	0.203342	0.830119	0.88675	0.972232	0.893961	0.0905135	0.899818
CFA	MobileNetV2	Capsules	0.230757	0.793232	0.6915	0.940224	0.789131	0.111665	0.792896
CFA	MobileNetV2	Cashew	0.618494	0.895826	0.9089	0.991864	0.95692	0.578428	0.820953
CFA	MobileNetV2	Chewinggum	0.10334	0.926601	0.9413	0.974413	0.974997	0.0467026	0.820782
CFA	MobileNetV2	Fryum	0.467768	0.884636	0.874	0.965602	0.938365	0.382852	0.910937
CFA	MobileNetV2	Macaroni1	0.0124868	0.714019	0.7401	0.959876	0.75398	0.00368207	0.874301
CFA	MobileNetV2	Macaroni2	0.00249828	0.723981	0.63055	0.852393	0.572981	0.000355292	0.754194
CFA	MobileNetV2	Pcb1	0.68592	0.869963	0.92985	0.995792	0.927636	0.742383	0.915668
CFA	MobileNetV2	Pcb2	0.0818268	0.844455	0.91015	0.979924	0.919044	0.0357068	0.906382
CFA	MobileNetV2	Pcb3	0.213318	0.867591	0.932921	0.989247	0.94137	0.124749	0.900683
CFA	MobileNetV2	Pcb4	0.354631	0.947889	0.981139	0.973996	0.975888	0.324655	0.835085
CFA	MobileNetV2	Pipe_fryum	0.607603	0.945813	0.9635	0.993713	0.980631	0.586515	0.919101
CFA	PhiNet	Candle	0.154736	0.824434	0.88865	0.957891	0.894639	0.0632806	0.864749
CFA	PhiNet	Capsules	0.208617	0.802877	0.726167	0.925765	0.800389	0.111756	0.803731
CFA	PhiNet	Cashew	0.657771	0.855676	0.849	0.993389	0.924214	0.654297	0.814601
CFA	PhiNet	Chewinggum	0.0612294	0.893454	0.9089	0.958683	0.957456	0.0251207	0.777164
CFA	PhiNet	Fryum	0.488497	0.866704	0.8766	0.972727	0.939484	0.412406	0.895083
CFA	PhiNet	Macaroni1	0.0079173	0.728835	0.74645	0.930474	0.727102	0.00142769	0.828528
CFA	PhiNet	Macaroni2	0.0174313	0.711296	0.6607	0.901293	0.620148	0.00111518	0.772873
CFA	PhiNet	Pcb1	0.498704	0.875622	0.9335	0.988402	0.929029	0.49966	0.870176
CFA	PhiNet	Pcb2	0.0660156	0.897373	0.93775	0.976657	0.946004	0.0255313	0.883142
CFA	PhiNet	Pcb3	0.179307	0.865608	0.93104	0.985183	0.937779	0.0894084	0.881964
CFA	PhiNet	Pcb4	0.331737	0.958843	0.990149	0.969516	0.990373	0.280603	0.821217
CFA	PhiNet	Pipe_fryum	0.653184	0.929101	0.9488	0.994561	0.975456	0.689868	0.85964
CFA	WideResNet50-2	Candle	0.232093	0.913333	0.9595	0.991481	0.963857	0.125767	0.958885
CFA	WideResNet50-2	Capsules	0.425742	0.80312	0.7645	0.987496	0.841818	0.35401	0.858221
CFA	WideResNet50-2	Cashew	0.708841	0.934	0.9566	0.996156	0.979352	0.749636	0.908357
CFA	WideResNet50-2	Chewinggum	0.229004	0.982561	0.9968	0.988082	0.998499	0.120574	0.873678
CFA	WideResNet50-2	Fryum	0.495592	0.930239	0.9466	0.967455	0.977074	0.412124	0.937095
CFA	WideResNet50-2	Macaroni1	0.0461246	0.834987	0.8861	0.989103	0.895397	0.00869534	0.952147
CFA	WideResNet50-2	Macaroni2	0.0229756	0.750255	0.7534	0.971645	0.74698	0.00235072	0.905322
CFA	WideResNet50-2	Pcb1	0.777521	0.953296	0.9864	0.998742	0.986084	0.862905	0.967421
CFA	WideResNet50-2	Pcb2	0.0934032	0.914538	0.9672	0.984833	0.972328	0.041764	0.927535
CFA	WideResNet50-2	Pcb3	0.245435	0.929993	0.969851	0.990336	0.972652	0.183031	0.906371
CFA	WideResNet50-2	Pcb4	0.372373	0.967907	0.994158	0.986566	0.994012	0.395643	0.890221
CFA	WideResNet50-2	Pipe_fryum	0.674519	0.99	0.995	0.995395	0.997606	0.694397	0.958648
PaDiM	MCUNet	Candle	0.0883304	0.786096	0.82885	0.975751	0.796978	0.0341335	0.887629
PaDiM	MCUNet	Capsules	0.0653379	0.77592	0.646583	0.888711	0.757595	0.0268542	0.633846
PaDiM	MCUNet	Cashew	0.548321	0.864031	0.8699	0.989551	0.936702	0.507946	0.812181
PaDiM	MCUNet	Chewinggum	0.189839	0.954769	0.9681	0.981688	0.985475	0.102198	0.875467
PaDiM	MCUNet	Fryum	0.377702	0.88736	0.8851	0.954191	0.930244	0.297434	0.831283
PaDiM	MCUNet	Macaroni1	0.00503883	0.762097	0.79	0.960324	0.748859	0.00172192	0.895493
PaDiM	MCUNet	Macaroni2	0.00196242	0.69937	0.60345	0.931691	0.579932	0.000647274	0.757428
PaDiM	MCUNet	Pcb1	0.372341	0.81636	0.8621	0.989425	0.850778	0.322045	0.911268
PaDiM	MCUNet	Pcb2	0.0558554	0.730153	0.76895	0.965875	0.772958	0.0233491	0.859192
PaDiM	MCUNet	Pcb3	0.120128	0.699902	0.671139	0.977014	0.674695	0.0791229	0.813348
PaDiM	MCUNet	Pcb4	0.196624	0.911184	0.948465	0.967217	0.930662	0.115015	0.822792
PaDiM	MCUNet	Pipe_fryum	0.573522	0.858407	0.7702	0.99051	0.864098	0.46182	0.906997
PaDiM	MicroNet-M1	Candle	0.00859069	0.724204	0.66465	0.667022	0.652432	0.00172527	0.599224
PaDiM	MicroNet-M1	Capsules	0.0226434	0.769231	0.425333	0.767313	0.603865	0.00737923	0.565361
PaDiM	MicroNet-M1	Cashew	0.284769	0.82358	0.6997	0.750532	0.821832	0.213621	0.593644
PaDiM	MicroNet-M1	Chewinggum	0.0068147	0.807307	0.5573	0.577858	0.743769	0.00247323	0.618695
PaDiM	MicroNet-M1	Fryum	0.192211	0.813145	0.685	0.900889	0.811663	0.138877	0.690474
PaDiM	MicroNet-M1	Macaroni1	0.00105471	0.689314	0.6184	0.888553	0.598322	0.00041534	0.595559
PaDiM	MicroNet-M1	Macaroni2	0.00170155	0.677314	0.50415	0.912144	0.497275	0.000490268	0.724088
PaDiM	MicroNet-M1	Pcb1	0.0452175	0.676914	0.53385	0.863096	0.546423	0.0169195	0.604772
PaDiM	MicroNet-M1	Pcb2	0.0286028	0.671312	0.4989	0.896799	0.499878	0.00867811	0.695154
PaDiM	MicroNet-M1	Pcb3	0.0352272	0.667781	0.520297	0.909905	0.529062	0.0129094	0.625771
PaDiM	MicroNet-M1	Pcb4	0.0244039	0.665559	0.491238	0.807371	0.488025	0.0106787	0.473629
PaDiM	MicroNet-M1	Pipe_fryum	0.284998	0.801606	0.4792	0.936572	0.690662	0.18949	0.653822
PaDiM	MobileNetV2	Candle	0.0943879	0.791101	0.83025	0.970198	0.787612	0.0323205	0.882386
PaDiM	MobileNetV2	Capsules	0.110005	0.775194	0.635833	0.927872	0.753816	0.042577	0.681973
PaDiM	MobileNetV2	Cashew	0.551394	0.86102	0.8655	0.986289	0.933091	0.480909	0.827529
PaDiM	MobileNetV2	Chewinggum	0.233428	0.951262	0.9727	0.984246	0.984955	0.12047	0.913041
PaDiM	MobileNetV2	Fryum	0.409751	0.880774	0.8679	0.956229	0.923027	0.306181	0.861359
PaDiM	MobileNetV2	Macaroni1	0.00427376	0.773357	0.8091	0.962542	0.787087	0.00157271	0.854273
PaDiM	MobileNetV2	Macaroni2	0.00210729	0.699202	0.64435	0.94402	0.630348	0.000800304	0.799464
PaDiM	MobileNetV2	Pcb1	0.4406	0.802722	0.8363	0.992304	0.818662	0.423801	0.916481
PaDiM	MobileNetV2	Pcb2	0.0846437	0.74099	0.77	0.978011	0.775989	0.0349651	0.891297
PaDiM	MobileNetV2	Pcb3	0.180088	0.732993	0.724208	0.984387	0.711686	0.106892	0.857697
PaDiM	MobileNetV2	Pcb4	0.205103	0.867928	0.891089	0.972566	0.856306	0.131883	0.842193
PaDiM	MobileNetV2	Pipe_fryum	0.542841	0.909946	0.9157	0.99008	0.952792	0.444526	0.930702
PaDiM	PhiNet	Candle	0.0966539	0.800832	0.8323	0.966983	0.790834	0.0327705	0.852545
PaDiM	PhiNet	Capsules	0.0493348	0.783461	0.63275	0.897427	0.747297	0.0189531	0.570241
PaDiM	PhiNet	Cashew	0.456019	0.87383	0.8912	0.983212	0.948926	0.385663	0.829896
PaDiM	PhiNet	Chewinggum	0.145004	0.941462	0.9679	0.97625	0.984326	0.0627043	0.821063
PaDiM	PhiNet	Fryum	0.337655	0.882391	0.8384	0.94374	0.906054	0.236786	0.813162
PaDiM	PhiNet	Macaroni1	0.00300588	0.801149	0.83895	0.95653	0.771986	0.00116003	0.841771
PaDiM	PhiNet	Macaroni2	0.00136395	0.700887	0.6307	0.908488	0.603587	0.000469551	0.692066
PaDiM	PhiNet	Pcb1	0.237318	0.836982	0.87745	0.97855	0.855598	0.137916	0.904613
PaDiM	PhiNet	Pcb2	0.0591043	0.763261	0.7724	0.972854	0.754049	0.0244828	0.86675
PaDiM	PhiNet	Pcb3	0.103618	0.729281	0.718218	0.977594	0.67335	0.0513754	0.836935
PaDiM	PhiNet	Pcb4	0.22171	0.897275	0.925891	0.97331	0.890364	0.125967	0.841796
PaDiM	PhiNet	Pipe_fryum	0.497322	0.902012	0.8869	0.987326	0.928208	0.382163	0.862103
PaDiM	WideResNet50-2	Candle	0.154664	0.80249	0.8315	0.985473	0.78533	0.0745679	0.945287
PaDiM	WideResNet50-2	Capsules	0.19594	0.773697	0.641417	0.968127	0.74931	0.0906467	0.766291
PaDiM	WideResNet50-2	Cashew	0.58389	0.881868	0.9119	0.991898	0.960034	0.556568	0.881356
PaDiM	WideResNet50-2	Chewinggum	0.29003	0.956923	0.9855	0.990975	0.993125	0.164269	0.914277
PaDiM	WideResNet50-2	Fryum	0.473087	0.89378	0.9055	0.968591	0.952305	0.392644	0.912784
PaDiM	WideResNet50-2	Macaroni1	0.0126835	0.800151	0.82305	0.984717	0.777115	0.00397682	0.942215
PaDiM	WideResNet50-2	Macaroni2	0.0195695	0.736947	0.67695	0.968608	0.622912	0.00219197	0.889085
PaDiM	WideResNet50-2	Pcb1	0.387177	0.863536	0.86615	0.993456	0.794478	0.313704	0.945077
PaDiM	WideResNet50-2	Pcb2	0.0900817	0.805807	0.86265	0.985488	0.860872	0.0462287	0.92871
PaDiM	WideResNet50-2	Pcb3	0.230062	0.724231	0.752079	0.989519	0.756654	0.178206	0.892216
PaDiM	WideResNet50-2	Pcb4	0.223715	0.905295	0.926782	0.973753	0.888509	0.13787	0.837998
PaDiM	WideResNet50-2	Pipe_fryum	0.61766	0.953741	0.9681	0.993583	0.982882	0.558135	0.964613
PatchCore	MCUNet	Candle	0.162357	0.850372	0.93455	0.936229	0.937346	0.0678308	0.851954
PatchCore	MCUNet	Capsules	0.216897	0.815521	0.82075	0.940403	0.904655	0.122722	0.781793
PatchCore	MCUNet	Cashew	0.309277	0.885119	0.9225	0.944647	0.964757	0.226596	0.805611
PatchCore	MCUNet	Chewinggum	0.235085	0.953124	0.9802	0.982835	0.990994	0.133071	0.849926
PatchCore	MCUNet	Fryum	0.310781	0.876339	0.911	0.916419	0.961231	0.24727	0.897476
PatchCore	MCUNet	Macaroni1	0.126988	0.856454	0.8928	0.966533	0.9084	0.0436615	0.890288
PatchCore	MCUNet	Macaroni2	0.0589528	0.70896	0.7322	0.922372	0.736514	0.00956059	0.890757
PatchCore	MCUNet	Pcb1	0.345703	0.933351	0.97635	0.976043	0.977957	0.270892	0.859398
PatchCore	MCUNet	Pcb2	0.161606	0.897857	0.94935	0.899986	0.960261	0.0662959	0.810172
PatchCore	MCUNet	Pcb3	0.256376	0.926933	0.974208	0.963595	0.974612	0.146994	0.856915
PatchCore	MCUNet	Pcb4	0.262141	0.977905	0.995	0.953777	0.994202	0.17959	0.788737
PatchCore	MCUNet	Pipe_fryum	0.477489	0.944787	0.9577	0.980379	0.975385	0.404565	0.910378
PatchCore	MicroNet-M1	Candle	0.0400892	0.728201	0.7365	0.714693	0.730337	0.00591783	0.665589
PatchCore	MicroNet-M1	Capsules	0.056918	0.777361	0.533333	0.778255	0.672099	0.0164499	0.663213
PatchCore	MicroNet-M1	Cashew	0.0943553	0.860341	0.8358	0.722293	0.902792	0.0378405	0.754639
PatchCore	MicroNet-M1	Chewinggum	0.0284374	0.808163	0.7181	0.648893	0.842289	0.00628367	0.696094
PatchCore	MicroNet-M1	Fryum	0.131391	0.809125	0.7398	0.758161	0.865402	0.0718963	0.781104
PatchCore	MicroNet-M1	Macaroni1	0.0035153	0.687658	0.6605	0.809394	0.649072	0.000563775	0.631335
PatchCore	MicroNet-M1	Macaroni2	0.00404042	0.671171	0.56695	0.864856	0.581272	0.000709882	0.660814
PatchCore	MicroNet-M1	Pcb1	0.123953	0.789215	0.83345	0.890031	0.809031	0.0442924	0.701508
PatchCore	MicroNet-M1	Pcb2	0.0118441	0.868223	0.9007	0.72593	0.924142	0.0021732	0.614531
PatchCore	MicroNet-M1	Pcb3	0.0340009	0.696994	0.710842	0.854433	0.749673	0.010364	0.681834
PatchCore	MicroNet-M1	Pcb4	0.0238542	0.673477	0.46703	0.730706	0.489003	0.00999761	0.517266
PatchCore	MicroNet-M1	Pipe_fryum	0.168507	0.818375	0.7166	0.882021	0.831406	0.0878602	0.757582
PatchCore	MobileNetV2	Candle	0.15729	0.854415	0.9133	0.924361	0.922879	0.0651537	0.845645
PatchCore	MobileNetV2	Capsules	0.278499	0.781667	0.734833	0.949601	0.840871	0.172132	0.771469
PatchCore	MobileNetV2	Cashew	0.277496	0.902498	0.9467	0.941451	0.975128	0.187784	0.810844
PatchCore	MobileNetV2	Chewinggum	0.168531	0.94324	0.973	0.957504	0.987883	0.0800977	0.771537
PatchCore	MobileNetV2	Fryum	0.250819	0.887454	0.9191	0.897578	0.961775	0.186726	0.862954
PatchCore	MobileNetV2	Macaroni1	0.0785967	0.763871	0.82785	0.934237	0.828412	0.0223983	0.853797
PatchCore	MobileNetV2	Macaroni2	0.0164502	0.694108	0.69825	0.873805	0.70604	0.00181647	0.799244
PatchCore	MobileNetV2	Pcb1	0.349245	0.915668	0.9724	0.975323	0.973487	0.277155	0.878244
PatchCore	MobileNetV2	Pcb2	0.164215	0.899757	0.9627	0.917595	0.968228	0.0689477	0.831774
PatchCore	MobileNetV2	Pcb3	0.237297	0.901817	0.959455	0.958306	0.957608	0.136339	0.826107
PatchCore	MobileNetV2	Pcb4	0.268697	0.987537	0.994703	0.958298	0.993269	0.195887	0.77327
PatchCore	MobileNetV2	Pipe_fryum	0.320569	0.954774	0.9677	0.962095	0.984729	0.224362	0.875198
PatchCore	PhiNet	Candle	0.107645	0.809307	0.87565	0.908568	0.882761	0.0330141	0.82641
PatchCore	PhiNet	Capsules	0.0911021	0.78483	0.684083	0.8515	0.819425	0.0333397	0.669122
PatchCore	PhiNet	Cashew	0.259716	0.84532	0.8575	0.94716	0.935818	0.156942	0.793279
PatchCore	PhiNet	Chewinggum	0.0959705	0.882109	0.8996	0.906984	0.954046	0.0321446	0.743392
PatchCore	PhiNet	Fryum	0.291451	0.841156	0.8617	0.937109	0.934654	0.218056	0.855826
PatchCore	PhiNet	Macaroni1	0.00507659	0.762407	0.80445	0.921943	0.797608	0.000904554	0.734784
PatchCore	PhiNet	Macaroni2	0.00237164	0.687393	0.5679	0.901072	0.536666	0.000507811	0.731659
PatchCore	PhiNet	Pcb1	0.142157	0.806598	0.8531	0.936151	0.852729	0.0716394	0.827358
PatchCore	PhiNet	Pcb2	0.0729903	0.81915	0.8779	0.942758	0.885036	0.0247695	0.822226
PatchCore	PhiNet	Pcb3	0.163846	0.779235	0.829356	0.96119	0.83494	0.0744533	0.803101
PatchCore	PhiNet	Pcb4	0.123847	0.941742	0.984604	0.942772	0.984207	0.0682427	0.717317
PatchCore	PhiNet	Pipe_fryum	0.34453	0.875576	0.8773	0.973856	0.94069	0.254775	0.824686
PatchCore	WideResNet50-2	Candle	0.182715	0.843133	0.92365	0.976655	0.929606	0.0925988	0.916186
PatchCore	WideResNet50-2	Capsules	0.240488	0.788694	0.72475	0.950765	0.835673	0.149341	0.786917
PatchCore	WideResNet50-2	Cashew	0.379893	0.929115	0.9545	0.958536	0.979091	0.285144	0.860884
PatchCore	WideResNet50-2	Chewinggum	0.258406	0.962063	0.9865	0.986577	0.994099	0.173371	0.860991
PatchCore	WideResNet50-2	Fryum	0.313711	0.878192	0.9132	0.923366	0.961992	0.240496	0.914054
PatchCore	WideResNet50-2	Macaroni1	0.0655239	0.789892	0.8521	0.964299	0.858386	0.017908	0.854323
PatchCore	WideResNet50-2	Macaroni2	0.0135284	0.68528	0.54255	0.944422	0.53675	0.00136429	0.830896
PatchCore	WideResNet50-2	Pcb1	0.373283	0.905495	0.94195	0.984207	0.937592	0.286756	0.908834
PatchCore	WideResNet50-2	Pcb2	0.135395	0.828502	0.9043	0.952364	0.915025	0.0610576	0.857689
PatchCore	WideResNet50-2	Pcb3	0.272728	0.834936	0.907327	0.971576	0.917527	0.157044	0.864959
PatchCore	WideResNet50-2	Pcb4	0.357987	0.952859	0.987723	0.976599	0.98751	0.280101	0.827027
PatchCore	WideResNet50-2	Pipe_fryum	0.410202	0.965541	0.9868	0.979862	0.993889	0.322881	0.941822
STFPM	MCUNet	Candle	0.115931	0.766575	0.78065	0.955312	0.727661	0.0431794	0.70553
STFPM	MCUNet	Capsules	0.26841	0.810386	0.82775	0.95471	0.899974	0.159025	0.919756
STFPM	MCUNet	Cashew	0.392583	0.84631	0.8336	0.853629	0.920811	0.315944	0.819955
STFPM	MCUNet	Chewinggum	0.271297	0.921491	0.9464	0.967494	0.977522	0.139451	0.914818
STFPM	MCUNet	Fryum	0.391239	0.907765	0.9145	0.928109	0.952374	0.326758	0.759174
STFPM	MCUNet	Macaroni1	0.0293812	0.72671	0.7545	0.99006	0.741678	0.0104184	0.85281
STFPM	MCUNet	Macaroni2	0.0236896	0.679301	0.5607	0.976046	0.52816	0.00457455	0.867744
STFPM	MCUNet	Pcb1	0.586595	0.804101	0.8457	0.992419	0.813581	0.602168	0.88183
STFPM	MCUNet	Pcb2	0.0581806	0.780028	0.80455	0.960341	0.787303	0.0248858	0.830142
STFPM	MCUNet	Pcb3	0.220567	0.69965	0.694455	0.977172	0.716942	0.120594	0.702358
STFPM	MCUNet	Pcb4	0.359284	0.930135	0.971089	0.982143	0.964032	0.264636	0.808966
STFPM	MCUNet	Pipe_fryum	0.575998	0.873056	0.8297	0.992198	0.901657	0.534022	0.773507
STFPM	MicroNet-M1	Candle	0.003985	0.72043	0.69265	0.545452	0.697773	0.00115389	0.522145
STFPM	MicroNet-M1	Capsules	0.00736437	0.793898	0.614083	0.594526	0.713912	0.00306033	0.511149
STFPM	MicroNet-M1	Cashew	0.0537349	0.8	0.4388	0.445509	0.651345	0.0207928	0.283201
STFPM	MicroNet-M1	Chewinggum	0.00362897	0.809046	0.5764	0.205807	0.712312	0.00105673	0.287783
STFPM	MicroNet-M1	Fryum	0.197031	0.808889	0.694	0.876826	0.813341	0.133947	0.652198
STFPM	MicroNet-M1	Macaroni1	0.00247357	0.672286	0.5221	0.703486	0.516892	0.00018918	0.522232
STFPM	MicroNet-M1	Macaroni2	0.00021909	0.67116	0.481	0.610249	0.486438	9.22286e-05	0.490054
STFPM	MicroNet-M1	Pcb1	0.0179485	0.685824	0.47345	0.637542	0.48262	0.00605427	0.470684
STFPM	MicroNet-M1	Pcb2	0.00563002	0.666667	0.45635	0.736243	0.499181	0.00189232	0.523414
STFPM	MicroNet-M1	Pcb3	0.0184176	0.674737	0.612327	0.832752	0.615116	0.00587209	0.557282
STFPM	MicroNet-M1	Pcb4	0.0104882	0.709348	0.71198	0.539217	0.683195	0.00469622	0.482023
STFPM	MicroNet-M1	Pipe_fryum	0.146786	0.803433	0.5752	0.842309	0.743801	0.0692609	0.538522
STFPM	MobileNetV2	Candle	0.134784	0.765601	0.77525	0.971924	0.718632	0.053001	0.69227
STFPM	MobileNetV2	Capsules	0.289019	0.806534	0.8155	0.971207	0.878691	0.171779	0.866541
STFPM	MobileNetV2	Cashew	0.457424	0.889423	0.9162	0.919452	0.960135	0.398164	0.815371
STFPM	MobileNetV2	Chewinggum	0.233495	0.907282	0.926	0.979383	0.969078	0.128545	0.886183
STFPM	MobileNetV2	Fryum	0.418271	0.885367	0.9054	0.943884	0.953565	0.377723	0.794035
STFPM	MobileNetV2	Macaroni1	0.0159746	0.747884	0.76495	0.990427	0.733323	0.00594644	0.649858
STFPM	MobileNetV2	Macaroni2	0.0127591	0.690693	0.57095	0.981467	0.518517	0.00407763	0.918475
STFPM	MobileNetV2	Pcb1	0.597614	0.819959	0.86465	0.993469	0.841576	0.64542	0.854905
STFPM	MobileNetV2	Pcb2	0.0670248	0.790821	0.80985	0.963965	0.799734	0.0284528	0.805507
STFPM	MobileNetV2	Pcb3	0.210261	0.779229	0.822376	0.981323	0.817854	0.127097	0.728579
STFPM	MobileNetV2	Pcb4	0.321206	0.929033	0.96297	0.9781	0.953341	0.215717	0.800361
STFPM	MobileNetV2	Pipe_fryum	0.576133	0.921951	0.9371	0.991525	0.965873	0.515362	0.792239
STFPM	PhiNet	Candle	0.050334	0.765578	0.76365	0.911823	0.696289	0.0166334	0.621496
STFPM	PhiNet	Capsules	0.251387	0.794071	0.778167	0.969592	0.859193	0.141336	0.787927
STFPM	PhiNet	Cashew	0.277815	0.828816	0.8264	0.772049	0.918812	0.195353	0.875675
STFPM	PhiNet	Chewinggum	0.133276	0.841684	0.8163	0.840136	0.899947	0.0447717	0.760824
STFPM	PhiNet	Fryum	0.377806	0.868472	0.8685	0.932648	0.93605	0.317763	0.882275
STFPM	PhiNet	Macaroni1	0.010109	0.736592	0.7567	0.974987	0.751974	0.00295703	0.760198
STFPM	PhiNet	Macaroni2	0.0108129	0.676028	0.50055	0.974644	0.504364	0.00285572	0.780706
STFPM	PhiNet	Pcb1	0.274547	0.706401	0.69635	0.961537	0.661173	0.156579	0.823578
STFPM	PhiNet	Pcb2	0.0581177	0.741694	0.7477	0.948058	0.713331	0.0235729	0.848767
STFPM	PhiNet	Pcb3	0.153792	0.722763	0.740495	0.963872	0.742467	0.0726276	0.670685
STFPM	PhiNet	Pcb4	0.16898	0.926447	0.965842	0.956876	0.960295	0.0972095	0.754459
STFPM	PhiNet	Pipe_fryum	0.481095	0.905385	0.9202	0.986929	0.961618	0.383801	0.80807