PruneTruong / DenseMatching

Dense matching library based on PyTorch
GNU Lesser General Public License v2.1
689 stars 81 forks source link

Help on understanding PDC-Net Figure3. #34

Open jinlovespho opened 1 month ago

jinlovespho commented 1 month ago
Screenshot 2024-10-25 at 11 46 19 AM

Hi thanks for the amazing work on PDC-Net.

I'm just having a hard time understanding Figure3 in PDC-Net.. and there was no where else to ask so I'm asking it here! This figure seems to be the motivation to use the mixture model (predicting M components of variances instead of one).

Can you give an easier explanation on figure3 ? especially on exactly what the inlier and outlier is in this context.

I think i would be able to better understand figure 3 with an example, so I provided you with an example below. Thanks in advance!!

(1) given two images, a query image Q and a reference image R, which have spatial size of (H,W). (2) GLUnet predicted the flowfield of R to Q as Y_pred. where Y_pred has shape of (H,W,2) (3) and we have the ground truth flow as Y_gt. where Y_gt has shape of (H,W,2) (4) now from the current setting, how was figure3 obtained?

what exactly does the x dimension ("Absolute Error (pixels)), inliers and outliers mean in figure 3? Thanks again!