erasaur / adversarial-patch-object-detection Goto Github PK

Hello, I would like to ask whether the method in the paper is just different from the untarget attack of DPatch in the update method, or what else I have not noticed. In addition, I would like to ask how to set the label for the untarge attack of DPatch. Isn't only set target_label as 0 the same as target? What about the paper that maximizes the object detection loss ignoring the loss of negative samples?

Hello，

Thanks for sharing the code, and your demo on youtube looks really cool.
We would like to checking on the idea in your paper, but we can't generate the similar results as you showed in your demo.
Now we are retrying the code with different configuration, could you provide some guidence on the key configuration/parameters which we can adjust them for better results? Thanks in advance!

For example, there is the following configuration in your code

    # maintain losses in a consistent order:
    # 1. all ground truth as usual
    # 2. ground truth for TARGET_CLASS removed
    # 3. all ground truth removed
    # 4. ground truth set to the patch with TARGET_CLASS (dpatch loss)
    "maximize_loss": [True, False, False, False],
    "target_loss_index": 3,
    "compute_all_losses": False,

• How shoud I choose from the 4 different options in the comments "maintain losses in a consistent order"?
• What's the meaning of [True, False, False, False] for "maximize_loss"？
• what is the meaning of 3 for "target_loss_index"? is it "3. all ground truth removed"?

Hello,

Thanks for sharing the source code for the great work, which can benefit a lot. I have a question about the results generated from the code.

First, I found some bugs from the jupyter code. For instance
# apply patch
tform_batch = get_random_tform_batch(len(image_batch))
Should it be
tform_batch = get_random_tform_batch(image_batch)
I think the input of this function should be images instead of a int number.

After fixing that, I found I cannot generate similar results reported in the paper. Mine is as follows (run for 5 times and select the best results). My mAP is very high (around 0.25). It looks very different compared to the zebra one in the paper and the notebook. I feel it may be because the randomness things for initialization or others. I am not very sure if there exists some other problems. Could you please help me on that? I think what I did is just change the path and the bugs I reported above, but can get a much worse results. Thanks a lot!