In the classification task, the predicted results of pixel segmentation are expected to give

[code10086web]

It is hoped that in the classification task, the prediction of pixel segmentation can be given when tested. What should I do?

[samet-akcay]

@code10086web, can you elaborate a bit please?

[code10086web]

For example， In a custom dataset, Since there is no pixel-level labelling, I set the task option to classification task instead of segmentation task. so during the inference phase, I can only get the result of whether the image is anomalous or not, and cannot know the result of where the anomalous area is roughly. But the predicted results of pixel-level are not given.

[code10086web]

Could you give me some tips? Thank you very much.

[ghost]

@code10086web You need to give a mask for the ground truth as far as I know. You could augment your dataset and create your own anomalous images and save a mask for the augmented area. I hope this is what you meant.

[code10086web]

Sorry for not being clear, I meant that even if there is no ground truth of the mask, I still want to get pixel-level prediction results as a reference.

[samet-akcay]

@code10086web, here is the link how the visualization images are generated. When there is no masks, anomalib treats it as a classification task, and only generates the predicted labels. You could potentially add

visualization.add_image(image_result.heat_map, "Predicted Heat Map")

to here in L125.

[code10086web]

Thanks a lot.

------------------ 原始邮件 ------------------ 发件人: "Samet @.>; 发送时间: 2022年8月15日(星期一) 晚上10:45 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [openvinotoolkit/anomalib] In the classification task, the predicted results of pixel segmentation are expected to give (Issue #489)

@code10086web, here is the link how the visualization images are generated. When there is no masks, anomalib treats it as a classification task, and only generates the predicted labels. You could potentially add visualization.add_image(image_result.heat_map, "Predicted Heat Map")
to here in L125.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

[samet-akcay]

Closing this issue, thinking it's resolved now. Feel free to re-open if there are further issues.