search

OpenDriveLab / PersFormer_3DLane

[ECCV 2022 Oral] Perspective Transformer on 3D Lane Detection
Apache License 2.0
426 stars 77 forks source link

how to choose top_view_region for custom dataset #106

Closed AnRanbel closed 1 year ago

AnRanbel commented 1 year ago

Hi,after i modify the (org_w,org_h) from (1920,1280) to (1640,590)，im_ipm = cv2.warpPerspective(im, H_im2ipm, (ipm_w, ipm_h)) got the black image, it seems related to top_view_region, i use the default value top_view_region = np.array([[-10, 103], [10, 103], [-10, 3], [10, 3]]). i want to know how to choose top_view_region for different scale dataset. looking forward to your reply, thank you.

ilnehc commented 1 year ago

The region is defined in the config file. But I think the issue might be with the transformation matrix.

AnRanbel commented 1 year ago

The region is defined in the config file. But I think the issue might be with the transformation matrix.

for openlane dataset, if (org_w,org_h)=1920,1280 work fine, but after change the scale ,output image will black. because i see in code the transformation matrix's computation used top_view_region, so I'm wondering if the value of "top_view_region" is critical to this problem.

ilnehc commented 1 year ago

I don't think the top_view region is critical. Did you adapt the intrinsic matrix accordingly?

AnRanbel commented 1 year ago

I don't think the top_view region is critical. Did you adapt the intrinsic matrix accordingly?

thank you for your reply, I found out that it was the camera parameters(extrinsic/intrinsic). I used the same parameters for a sample in the openlane dataset, which didn't work on my dataset.