search

LeeDoYup / AnoGAN-tf

Unofficial Tensorflow Implementation of AnoGAN (Anomaly GAN)
MIT License
270 stars 92 forks source link

anomaly score #19

Open lzzlxxlsz opened 5 years ago

lzzlxxlsz commented 5 years ago

When I finished the anomaly test stage with the anomal score,and how can I know the anomaly picture

LeeDoYup commented 5 years ago

In this paper, the detail method is not presented. However, the basic approach is construct validation dataset and set anomaly threshold based on validation sets.

lzzlxxlsz commented 5 years ago

Thank you! That means I have to set a threshold to determine whether a picture is abnormal or not.With your test_data_01.png ,the anomal score is 3185.73901367,is result biggger？Thanks a lot !

LeeDoYup commented 5 years ago

There is anomaly score in each pixel and the value is the aggregation of all the pixel. So it can be large. Rather than considering absolute value, you can define the threshold based on validation detection accuracy. (That is, you need labeled-anomaly-data for validation)

lzzlxxlsz commented 5 years ago

That is to say,I use the validation dataset(labeled) for testing,then find a better threshold

lzzlxxlsz commented 5 years ago

There is another question ，what is the anomal situation for the celebA dataset？is there a hat or glasses and be the anomal ?

pradeepbhaskar commented 4 years ago

Hi Lee,

Could you please elaborate what does "labeled-anomaly-data" mean. When I am testing my model using anomaly image, I get 2 images on the "test_results" directory. First image file is all blank and other mage file ( which has "error" in its name) is the gray scale image of the anomaly image. A bit confused as to how to interpret.

Epoch: [97], anomaly score: 71413.89062500, res loss: 68573.62500000 Epoch: [98], anomaly score: 71413.28906250, res loss: 68573.62500000 Epoch: [99], anomaly score: 71412.67187500, res loss: 68573.60937500

./test_data/test_result/AD_0_IMG_20191020_120417.jpg ./test_data/test_result/AD_50_IMG_20191020_120417.jpg ./test_data/test_result/AD_99_IMG_20191020_120417.jpg

./test_data/test_result/AD_error_0_IMG_20191020_120417.jpg ./test_data/test_result/AD_error_50_IMG_20191020_120417.jpg ./test_data/test_result/AD_error_99_IMG_20191020_120417.jpg

Right now I am training with 100 epochs, should I train more ? My test epoch is 100 , should it be increased ? Please assist.

thanks, Pradeep

sachitha-bandara commented 4 years ago

@pradeepbhaskar were you able to find the answers for the above question? @LeeDoYup I am also confused with the images saved when the test is running. Please elaborate. Thanks in advance.