LoFTR
LoFTR copied to clipboard
zju3dv
Code for "LoFTR: Detector-Free Local Feature Matching with Transformers", CVPR 2021, T-PAMI 2022
Hi, I am trying to use LoFTR in my own project, while I find that I get different matching results when changing the input order of image pair. Is this...
Hi, thanks for your excellent work，it's really impressive. I have a question. It is mentioned that you select image pairs having overlap from 40% to 80% for training. Is the...
Hi, I replase your 128 dimension feature map; in resnet_fpn.py 'x0 = self.relu(self.bn1(self.conv1(x)))'; i want to use my cnn net, but the result is different every time , sometimes matches...
I see in the paper the authors say they trained LoFTR on Scannet with images resized to 480 x 640 pixels, and on Megadepth with images resized to have the...
Due to equipment problems, during the experiment，my trainval's DATASET.MGDPT_IMG_RESIZE, once set to 640, the GPU video memory will be insufficient and I have to reduce it. I want to know...
loss=nan
When I tried to train on a partial scannet dataset (contructed as you guys did), sometimes the logging shows that "loss =nan". I loaded the checkpoint you provided. Note that...
Is it possible to use LOFTR for the image classification task? How can I prepare data for the training classification task?
Did you provide the weight file of ResNetFPN_16_4? Or do we need to train ourselves？ Thanks!
Thanks for your great work! In supplementary material, instead of Post-LN layer, you propose an encoder layer variant which has fast convergence without extensive hyperparameter tuning. The difference of your...
Hi. Thank you for your great work. By the way, [the following line](https://github.com/zju3dv/LoFTR/blob/5d6c83428ab57987e5c4d42374b86e8b1f9cb520/src/utils/metrics.py#L80) may be a tiny bug that only affect thresholds for the RANSAC algorithm. ```` ransac_thr = thresh...
