Visualization

[sheshap]

Hi,

Thank you for sharing the code. Can you please provide instructions for the visualization results?

Thanks

[LavenderLA]

Hi, thanks for your interest in our work!

You can visualize the colorful projected image via the img tensor in this line: https://github.com/wangzy22/P2P/blob/13169a3029e0706b9de4d20b98662181699fc7d3/models/layers/encoder.py#L106

[auniquesun]

Hi, thanks for your interest in our work!

You can visualize the colorful projected image via the img tensor in this line:

https://github.com/wangzy22/P2P/blob/13169a3029e0706b9de4d20b98662181699fc7d3/models/layers/encoder.py#L106

Hi, author, thanks for sharing the paper and code!

I run the code you mentioned when testing with config p2p_ConvNeXt-B-1k on ModelNet40, and save all img in a batch (64 images) with opencv-python successfully, as the code shows.

@staticmethod
def save_imgs(imgs):
    imgs = imgs.permute(0, 2, 3, 1)
    for i, img in enumerate(imgs):
        img = img.cpu().numpy()
        cv.imwrite(f'./img_{i}.png', img)

However, I found all saved images are just black background, an example is presented as below

The results are obviously different from the projection visualization presented in the paper, what's the possible problem and can you give some insights or advice?

[LavenderLA]

Hi @auniquesun, thanks for your interest in our work!

We use to_pil_image function in torchvision to visualize our image. For example,

def vis_img(self, img):
    import torchvision.transforms.functional as tvF
    pic = tvF.to_pil_image(img)
    pic.save('vis_img.jpg')

Then you can use this function to visualize image:

img = self.img_layer(img_feat)  # B 3 224 224
self.vis_img(img[0].detach().cpu())

[auniquesun]

Thank you, I got it.