Human-Path-Prediction
Human-Path-Prediction copied to clipboard
HarshayuGirase
ETH original dataset
Thank you very much for your work, I have a request I hope you can agree to Could you share with me the data of your ETH processing script before processing, thank you very much!
hello, I have the same question. I can't get the same *.pickle data from original data(sgan provided) using image2world function, do you have some progress?
你好，我有同样的问题。我无法使用 image2world 函数从原始数据（提供 sgan）中获取相同的 *.pickle 数据，您有什么进展吗？
Hello, I'm sorry to tell you that I haven't made any progress on this issue, so I turned to other model explorations. If you have any questions, please contact me by email.
Hello, I'm sorry to tell you that I haven't made any progress on this issue, so I turned to other model explorations. If you have any questions, please contact me by email.
Hello, now I know how to make the world-pixel transformation using the homography matrix, but I still don't know how Ynet filter the data, so I turned to other model explorations, too. Here's the world2image transformation(from ETH official guidance), I hope it can help someone who are concerned about this issue:
def world2image(traj_w, H_inv):
# Converts points from Euclidean to homogeneous space, by (x, y) \u2192 (x, y, 1)
traj_homog = np.hstack((traj_w, np.ones((traj_w.shape[0], 1)))).T
# to camera frame
traj_cam = np.matmul(H_inv, traj_homog)
# to pixel coords
traj_uvz = np.transpose(traj_cam/traj_cam[2])
return traj_uvz[:, :2]
Hello, I'm sorry to tell you that I haven't made any progress on this issue, so I turned to other model explorations. If you have any questions, please contact me by email.
Hello, now I know how to make the world-pixel transformation using the homography matrix, but I still don't know how Ynet filter the data, so I turned to other model explorations, too. Here's the world2image transformation(from ETH official guidance), I hope it can help someone who are concerned about this issue:
def world2image(traj_w, H_inv): # Converts points from Euclidean to homogeneous space, by (x, y) \u2192 (x, y, 1) traj_homog = np.hstack((traj_w, np.ones((traj_w.shape[0], 1)))).T # to camera frame traj_cam = np.matmul(H_inv, traj_homog) # to pixel coords traj_uvz = np.transpose(traj_cam/traj_cam[2]) return traj_uvz[:, :2]
I would like to ask if the UCY dataset has the same homography matrix for converting coordinates between world coordinates and pixel coordinates. Similar work, Y-net, NSP-SFM, etc., all use map information in pixel space. The final indicators (ADE/FDE) are in world coordinates. How do they achieve conversion on UCY? Also I noticed that the original dataset of UCY seems to be in pixel coordinates, most of the existing work uses world coordinates, how is this converted, thank you very much
