rocketflash / cap_augmentation Goto Github PK

Hi, @RocketFlash ,

• Thanks for making this up, can u please tell me how to integrate cut-paste with detectron2 augmentations.

Hey, Nice Project. I was wondering about your implementation for generating birds eye view, will it be possible to provide work/links/documents that you referred for this conversion ? which can help me understand your implementation. I tried your code on my camera parameters and I get a black Image.

Regards

hey, first of all, great work, thank you.
I am getting this error;

ValueError: empty range for randrange() (0,0, 0)

which should mean there's a problem with the randint() function, mostly that in randint(a,b), b>a which somehow isn't the case here. can you help me with this?

I’d like to paste the segmented object (e.g. person) onto the target image, similar to your demo, instead of the whole image that contains the person.

Below is my figure:


The result is as follows:

Hello, I want konw you batch_size and input image_size

Thanks for your great work.

But I encountered some problems when I run the code.
To be specific, there seems to be something wrong about the semantic_mask and the instance_mask.
The code can't draw instance(or semantic) segmentation image correctly.

I can't understand the logic stream in these codes above.

I try to modify them to the following:

image_dst, coords, mask = self.paste_object(image_dst, image_src, x_coord, y_coord, self.random_h_flip, self.random_v_flip)
if coords:
    coords_all.append(coords)
    x1, y1, x2, y2 = coords
    semantic_mask[y1:y2, x1:x2] = mask
    instance_mask[y1:y2, x1:x2] = mask*(idx+1)

And it's actually works.

So, I want to know more about the code snippet above. Can it work rensonably？