edosedgar / mtcnnattack Goto Github PK

Hello
First, thank you for that wonderful contribution.

1. What is this checkerboard for?
   Why did you use it instead of just marking the 8 points straight on the face? (with no checkerboard at all).
2. Does your code generate a mask on the checkerboard's place, or uses it as a physical mask? (in order to apply the patch addition on the image).
   Thank you!

Thank you for your good work, Why not select RGB and how to generate RGB？

@edosedgar I tried training with 6 images and for 2000, 4000 epochs, but the patch trained seems of no use in detection using MTCNN. Even with trained patches, the detector can detect faces.

Hi Dear,
thank you for sharing your work,
I tested ur code and is working fine, however I could not figure out how to set my own mask on my photo.
I would be grateful if you explain the steps.
Regards

Hello！

I'm a third year college student, and I'm learning about the project of adversarial attack. This is the first project I studied.

Due to the version problem of tensorflow and python. I have been debugging problems with the version of the software for a long time.

So would you like to tell me the version of tensorflow and python in this project that you use?

Hello.
I've changed the input photo with my own photo with two checkerboard squares on my face.
Then when I run adversarial_pen.py,there is a error:
`
File ".\mtcnnattack-master\utils\patch_mng.py", line 124, in init_src_parts
x = np.linspace(0, self.mask_np.shape[1], self.grid_size[1]).astype(np.int32)

TypeError: 'NoneType' object is not subscriptable`

If the reason is that I didn't set masks = {
'left_cheek: [np zeros ((150, 180, 1)), (0, 255, 0), (0, 1, 0)].
'right_cheek: [np zeros ((150, 180, 1)), (255, 0, 0), (1, 0, 0)].
}

And How can I set the masks?
Thanks a lot!

Hello
First of all, thank you for your great contribution.

I am curious how you will verify the effect of the "attacked1.png" you generate.
I have tried detect_face(attacked_face) but there are no output.

I fixed it by this repo.
https://github.com/Polkach/cv-inter_area-on-tensorflow

Originally posted by @guoth in #8 (comment)

Did you succeed in the end?

I put inter_ Replace the area code with the Inter in the link you provide_ Area, no confrontation patch can be generated in the digital domain

How can I get my own input_ img? and I don't quite understand how mask works，please.

What do the three parameters in this line mean? I found that the latter is used to create PNet for each size and calc PNet probability map loss, but I don't quite understand how to adjust this to suit my own solution.

adv_mask.set_input_sizes([(73, 129), (103, 182), (52, 92)])

Anyway, THX!

Hello, I'm having the following error in your code:
"AttributeError: module 'tensorflow' has no attribute 'contrib'"
Yes, I know this is because of the TensorFlow older version.
But can you please suggest the following code modification for the TensorFlow 2.3 or latest version?

patch_transformed = tf.contrib.image.transform(patch_batched, Mb,
interpolation="NEAREST",
name="perspective")

N.B.: I know that "tf.contrib" has moved out of TF starting TF 2.0 alpha.
Thanks in advance.

@edosedgar

Hello.
I've changed the input photo with my own photo with two checkerboard squares on my face.

The grid_size with your image is (3,3) but with my image its (1,9).

so, I am getting this error

Traceback (most recent call last):
File "adversarial_gen.py", line 266, in
adv_mask.build(sess)
File "adversarial_gen.py", line 57, in build
self.pm.compile()
File "/ssd_scratch/cvit/surabhi/mtcnnattack/utils/patch_mng.py", line 188, in compile
self.patches[patch_key].init_transforms(self.images)
File "/ssd_scratch/cvit/surabhi/mtcnnattack/utils/patch_mng.py", line 148, in init_transforms
xx_dst, yy_dst = image.get_grid(self.init_clr, self.grad_clr, self.grid_size)
File "/ssd_scratch/cvit/surabhi/mtcnnattack/utils/patch_mng.py", line 62, in get_grid
xx_dst, yy_dst = self.add_new_line(xx_dst, yy_dst, dots_mask, i)
File "/ssd_scratch/cvit/surabhi/mtcnnattack/utils/patch_mng.py", line 52, in add_new_line
xx[i,:], yy[i,:] = points[:,1], points[:,0]
ValueError: could not broadcast input array from shape (20) into shape (9)

Dear sir, would you mind share the code realizing the projective transform? I have not found the codes and the input images of adversarial_gen.py are the images with patches which has been projected to the face. Thank you sir.

@edosedgar It would be really helpful if you can explain the steps of the pipeline in short. After applying checkerboard on the face, how projective transform is done ? I didn't get the concept of patch. Is patch the checkerboard or something else ?