Feature Extraction
Closed this issue · 6 comments
Hello,author, I would like to inquire why feature extraction is often performed on single-channel gray images. What are the advantages of this approach and is it possible to use three-channel color images for feature extraction?
Hi @jia199903, I think there is no special advantage when using gray images. If you train your model with your dataset, I recommend using RGB images instead. The main reason why some methods use gray images is because of evaluation on the HPatches dataset (this dataset only includes gray images).
Hello,author, @TruongKhang ，I want to know if this method can be used for image registration or image stitching tasks？I look forward to your reply.
It depends on whether the two tasks require feature matching or not. I'm not an expert in those so I can't tell you in detail.
@TruongKhang
Thank you very much for your response！
if this features here are already enhanced features by Topic-aware,Can I directly use the features obtained here to do correlation calculations？
Hi @jia199903 , yes, the features feat_c0 and feat_c1 are enhanced using topics. But remember that these features are extracted at low-resolution images (H/8 x W/8)