search

flyinglynx / Bilinear-Matching-Network

Official implementation for CVPR 2022 paper "Represent, Compare, and Learn: A Similarity-Aware Framework for Class-Agnostic Counting".
MIT License
69 stars 9 forks source link

image size and density map size #3

Closed trungpham2606 closed 2 years ago

trungpham2606 commented 2 years ago

Dear @flyinglynx In your dataloader code, I see that img will be resized (in a range), but I dont see where you change the density map's size as well ? Do I miss anything here :-?

flyinglynx commented 2 years ago

Hi! Thank u, that's an error in our code. But it will not trigger any error during training. Because all the images in FSC-147 have already satisfied the size requirement. Maybe authors of FSC-147 have preprocessed the dataset. We will fix this issue, and you can implement a simple interpolation (opencv or torch have some functions to do so) to resize the density maps as well.

trungpham2606 commented 2 years ago

@flyinglynx I see ^^. thank you Btw, do you have the code to generate gt density map ?

flyinglynx commented 2 years ago

Sorry, I used the provided density maps. I see some relevant issues in "https://github.com/cvlab-stonybrook/LearningToCountEverything". Maybe their discussion can help you.