You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Thanks for sharing the code! I'm able to run the latest version (the forked repo from bradyz) and trying to generate optical flow data for training.
However, I find that flow on human limbs are usually incorrect. If I warp the prev. frame according to the flow, some parts of the human are missing while pixels of the cars and background are aligned well. As shown in the figure below (left: current frame, right: warped previous frame), the leg region circled in red in missing.
To further examine this issue, I plot the 3D points of the (t-1) frame in red and the 3D points in the prev_pos variable of the (t) frame in blue under the same coordinate. They are supposed to perfectly overlap. However, the points on the limbs do not overlap.
I'm trying to debug this, but find it difficult. Could you give me some hints of what could go wrong?
The text was updated successfully, but these errors were encountered:
Thanks for sharing the code! I'm able to run the latest version (the forked repo from bradyz) and trying to generate optical flow data for training.
However, I find that flow on human limbs are usually incorrect. If I warp the prev. frame according to the flow, some parts of the human are missing while pixels of the cars and background are aligned well. As shown in the figure below (left: current frame, right: warped previous frame), the leg region circled in red in missing.
To further examine this issue, I plot the 3D points of the (t-1) frame in red and the 3D points in the prev_pos variable of the (t) frame in blue under the same coordinate. They are supposed to perfectly overlap. However, the points on the limbs do not overlap.
I'm trying to debug this, but find it difficult. Could you give me some hints of what could go wrong?
The text was updated successfully, but these errors were encountered: