Performances per category in the MVTec LOCO

[mjack3]

Hello,

I want to congratulate you for this great work. I wonder if you have the performance per subset of MVTec LOCO data in structural and logical terms. For example, what is the structural and logical performance of the "breakfast box" data subset?
Regards.

[hmyao22]

Hello,

I want to congratulate you for this great work. I wonder if you have the performance per subset of MVTec LOCO data in structural and logical terms. For example, what is the structural and logical performance of the "breakfast box" data subset? Regards.

Thank you for your attention to this work. Per-class results on the Mvtec LOCO dataset are available in the full accepted article.(H. Yao et al., "Dual-Attention Transformer and Discriminative Flow for Industrial Visual Anomaly Detection," in IEEE Transactions on Automation Science and Engineering, doi: 10.1109/TASE.2023.3322156. )

[mjack3]

Hi,I mean per-category and type of anomaly. For example, what is the structural and logical performance of the "breakfast box" data subset?

[hmyao22]

Hi,I mean per-category and type of anomaly. For example, what is the structural and logical performance of the "breakfast box" data subset?

Hey, I understand what you mean bro. Here are the results:
Breakfast box :
"classification": {
"auc_roc": {
"logical_anomalies": 0.7577368296716278,
"mean": 0.7528335564480143,
"structural_anomalies": 0.7479302832244008
}
},
"localization": {
"auc_spro": {
"logical_anomalies": {
"0.01": 0.23851688165194795,
"0.05": 0.42030049268785186,
"0.1": 0.5171552883706604,
"0.3": 0.6932737185740211,
"1.0": 0.8876883259847754
},
"mean": {
"0.01": 0.30339016998155405,
"0.05": 0.5378453272294377,
"0.1": 0.6309044485678232,
"0.3": 0.7742798907730943,
"1.0": 0.9186257028896639
},
"structural_anomalies": {
"0.01": 0.36826345831116014,
"0.05": 0.6553901617710235,
"0.1": 0.744653608764986,
"0.3": 0.8552860629721676,
"1.0": 0.9495630797945525
}
},

Bottle:
{
"classification": {
"auc_roc": {
"logical_anomalies": 0.9870392568175007,
"mean": 0.9857672517677363,
"structural_anomalies": 0.984495246717972
}
},
"localization": {
"auc_spro": {
"logical_anomalies": {
"0.01": 0.7045757602056713,
"0.05": 0.8925472523245211,
"0.1": 0.9444198073663437,
"0.3": 0.9814014697440837,
"1.0": 0.994420440923225
},
"mean": {
"0.01": 0.6788655484650101,
"0.05": 0.871671292865424,
"0.1": 0.9218357518363609,
"0.3": 0.9681944837869352,
"1.0": 0.9900926861636277
},
"structural_anomalies": {
"0.01": 0.6531553367243489,
"0.05": 0.8507953334063268,
"0.1": 0.8992516963063782,
"0.3": 0.9549874978297866,
"1.0": 0.9857649314040302
}
},
Pushpins:
{
"classification": {
"auc_roc": {
"logical_anomalies": 0.7666825927695493,
"mean": 0.8100723752897666,
"structural_anomalies": 0.8534621578099839
}
},
"localization": {
"auc_spro": {
"logical_anomalies": {
"0.01": 0.34047545512150906,
"0.05": 0.6517577880415395,
"0.1": 0.7670278940137965,
"0.3": 0.9065423178191194,
"1.0": 0.9719626953457359
},
"mean": {
"0.01": 0.40955201161586363,
"0.05": 0.6637830423515244,
"0.1": 0.7601456663276573,
"0.3": 0.8913403497944923,
"1.0": 0.9651557837533352
},
"structural_anomalies": {
"0.01": 0.4786285681102182,
"0.05": 0.6758082966615093,
"0.1": 0.7532634386415182,
"0.3": 0.8761383817698652,
"1.0": 0.9583488721609346
}
},
Screw bag:
"classification": {
"auc_roc": {
"logical_anomalies": 0.6623788440828049,
"mean": 0.7731126527491193,
"structural_anomalies": 0.8838464614154338
}
},
"localization": {
"auc_spro": {
"logical_anomalies": {
"0.01": 0.12210382663195764,
"0.05": 0.3753745777212909,
"0.1": 0.503156157102832,
"0.3": 0.7438426171030394,
"1.0": 0.9193061257459721
},
"mean": {
"0.01": 0.3080911160464428,
"0.05": 0.570607389633623,
"0.1": 0.6783363080090267,
"0.3": 0.8437802984623054,
"1.0": 0.95120238769088
},
"structural_anomalies": {
"0.01": 0.4940784054609279,
"0.05": 0.765840201545955,
"0.1": 0.8535164589152213,
"0.3": 0.9437179798215714,
"1.0": 0.9830986496357876
}
},
Connector:
{
"classification": {
"auc_roc": {
"logical_anomalies": 0.7848583877995643,
"mean": 0.8634128814924663,
"structural_anomalies": 0.9419673751853683
}
},
"localization": {
"auc_spro": {
"logical_anomalies": {
"0.01": 0.3972030896423277,
"0.05": 0.622604697233868,
"0.1": 0.739499826907959,
"0.3": 0.8970071288936207,
"1.0": 0.9683996637207837
},
"mean": {
"0.01": 0.49607379173534916,
"0.05": 0.7244507585255396,
"0.1": 0.8127777381850738,
"0.3": 0.9245598768802799,
"1.0": 0.976937182344492
},
"structural_anomalies": {
"0.01": 0.5949444938283706,
"0.05": 0.8262968198172113,
"0.1": 0.8860556494621887,
"0.3": 0.9521126248669392,
"1.0": 0.9854747009682003
}
},

[hmyao22]

Hi,I mean per-category and type of anomaly. For example, what is the structural and logical performance of the "breakfast box" data subset?

By the way, my WeChat number is 19911813371, feel free to contact me if any trouble

[mjack3]

Thank you very much!

[mjack3]

Hello, I hope you are doing well.

Could you tell me how I can launch this repository to train its method in the MVTec AD? To simplify things, let's assume that the dataset is at the root of the project. I would like to reproduce some anomaly maps to do a qualitative analysis.

Regards.