mattpoggi / depthstillation

Demo code for paper "Learning optical flow from still images", CVPR 2021.
MIT License
152 stars 11 forks source link

Object motion in the generated dataset. #9

Closed Sharpiless closed 1 year ago

Sharpiless commented 1 year ago

Thanks for your great work! But I have a small question about the moving object in the paper. Table 1 shows that adding moving objects gives better performance. However, in Table 2 and the following tables, the dCOCO does not appear to contain moving objects (with EPE=3.81 on KITTI 15). Do you implement the following experiments, including dDAVIS and dKITTI, without moving objects? This is very important to us because our recent work uses depthstillation as an important reference. We would appreciate receiving your reply.

mattpoggi commented 1 year ago

Hi, sorry for being late. The experiments after Tab. 2 are WITHOUT moving objects, as stated at the bottom of page 6, left column: "Thus, since our primary focus is on real environments, we choose (C) as the configuration for the following experiments."

Sharpiless commented 1 year ago

Thanks!