haofanwang / Score-CAM

Official implementation of Score-CAM in PyTorch
MIT License
399 stars 66 forks source link

Point Game Function #23

Closed RuoyuChen10 closed 2 years ago

RuoyuChen10 commented 2 years ago

Would you please offer your Point Game code for an evaluation? Thanks.

haofanwang commented 2 years ago

@RuoyuChen10

The implementation of Point Game should be easy. It works by checking whether the maximum point of map falls into the bounding box. I have provided Energy-based Point Game for your reference, slight modification should be fine.

RuoyuChen10 commented 2 years ago

@RuoyuChen10

The implementation of Point Game should be easy. It works by checking whether the maximum point of map falls into the bounding box. I have provided Energy-based Point Game for your reference, slight modification should be fine.

Thanks for your reply, and I also have a question, if there are multi objects in the model, how can I quantity the PG metric? For example, in the classification task, there are two cats in the image and two bounding boxes, if the maximum point of the map falls into one of the boxes it's ok? Thanks.

haofanwang commented 2 years ago

For multi-objects localization, I would suggest you adopt our Energy-based Point Game instead of Point Game. Otherwise, the score of Point Game will be inflated as the maximum has higher possibility falling into bbox.