search

DarioFontanel / PAnS

Detecting Anomalies in Semantic Segmentation with Prototypes
MIT License
24 stars 3 forks source link

The color of the Anomaly Segmentation pictures #7

Closed JackyLin7414 closed 2 years ago

JackyLin7414 commented 2 years ago

Hi @DarioFontanel , I added this line like you said in https://github.com/DarioFontanel/PAnS/issues/5#issuecomment-1079902994 plt.imshow(anomaly_probabilities.squeeze(0).cpu().numpy(), interpolation='bilinear', cmap=matplotlib.cm.Blues, vmin=-1, vmax=1)

to here https://github.com/DarioFontanel/PAnS/blob/fb604d01f9de57e83b2adfddf60e7a9fa6615bc6/train.py#L206

But I get an image like this as

Color is not the same image

Can you tell me why?

Thank you. I appreciate it.

DarioFontanel commented 2 years ago

Hi, It might simply be due to scale factor multiplication. Let's try different options of vmin and vmax factors to increase/decrease the normalization (I would advise starting with 0.5) and see if this is the cause.

JackyLin7414 commented 2 years ago

Hi, @DarioFontanel Can you tell me which image number you chose?

Thank you. I appreciate it.

Jacky

DarioFontanel commented 2 years ago

Hi, This list might contain some of the images reported in the final version of the paper, but I am not 100% sure because unfortunately I can't access the notebook updated version anymore. list_idx = [475, 798, 998, 1024].

Let me know if that works.

Dario

JackyLin7414 commented 2 years ago

Hi @DarioFontanel ,

The number of Anomaly Segmentation pictures is five instead of four. I tried the list you provided. But all the images I get are different from yours.

Thank you. I appreciate it.

Jacky