search

qianqianwang68 / caps

MIT License
183 stars 25 forks source link

Problem with grayscale images #5

Closed jeethesh-pai closed 3 years ago

jeethesh-pai commented 3 years ago

Hallo caps team,

I have a doubt on drawing inference on images which are grayscale. Is it possible to get feature descriptor using CAPS network for a grayscale image. I get an error stating the pretrained model weights are incompatible for 1-channel inputs ?

Thanks Jeethesh

qianqianwang68 commented 3 years ago

Hi you can try to turn the grayscale image into a color image by stacking the grayscale image three times along the color channel. Then you can feed the output image into the our network. Hope this helps.

jeethesh-pai commented 3 years ago

Hi, Thank you for your quicl reply. I tried that but the match is not optimal. Many keypoints are not matched. Is there a way to train with grayscale image so that i can increase the correspondence match?

qianqianwang68 commented 3 years ago

Yes, you can turn the color images in the provided dataset into gray-scale images and retrain/finetune the model.

jeethesh-pai commented 3 years ago

Thank you so much replies. I will try training with grayscale images.