search

Megum1 / LOTUS

[CVPR'24] LOTUS: Evasive and Resilient Backdoor Attacks through Sub-Partitioning
https://www.cs.purdue.edu/homes/cheng535/
MIT License
8 stars 0 forks source link

some confusion #2

Open JoyRiderJie opened 6 days ago

JoyRiderJie commented 6 days ago

Hi, thank you for sharing!LOTUS is a very interesting work.

When I read the "test" section of the "train. py" file,I was a bit confused about the data processing code for the second half of the function prediction. Can you share the writing ideas for this part of the code?

I am looking forward to hearing your reply.

Megum1 commented 3 hours ago

Hi. Thanks for your interest in our work!

Apologies for the confusion earlier. To address your question, we have uploaded a new file, test_partition.py, which includes detailed explanations of our testing process.

In essence, we compute the ASR and ASR_par for each image by applying different trigger combinations. These combinations are represented using binary codes. For instance, "0101" indicates that the 2nd and 4th triggers are applied to the input image.

We hope this clarifies our approach and helps you better understand our work.