search

DRosemei / RoMe

214 stars 26 forks source link

problems when using custom dataset #17

Closed yongguanjiangshan closed 12 months ago

yongguanjiangshan commented 12 months ago

hi, @DRosemei , thank you for your advances and patience. now i can use my custom dataset to generate bev_rgb and bev_seg. but now the blend image is bad. could you give me some suggestions to solve this? below are some result images.

bev_rgb: bev_rgb_20_1fc1cc6e05b61034d48b

bev_seg: bev_seg_20_b87ac3a2b3c37afe30fd

blend_image: blend_image_20_0ede2b3197a6b42b08cf

gt_image: gt_image_20_efb25c6fe6f5163d1f55

render_gt_seg: render_gt_seg_20_04906dcb1e4cfa3c3a56

render_image: render_image_20_0cf65de6a996fb3c1876

render_mask: render_mask_20_5c3ad3468872285bc206

render_seg: render_seg_20_d7c7ae149935f8428c44

DRosemei commented 12 months ago

@yongguanjiangshan I find that your images have not been undistored and you can rectify them first. image

yongguanjiangshan commented 12 months ago

@yongguanjiangshan I find that your images have not been undistored and you can rectify them first. image

hi, DRosemei, i have rectify the image. my original image is a fisheye image. as show below. By the way, i use the rectify algorithm in our company to rectify the image instead of the undistort() function in opencv. is it the rectify algorithm problem?

original fisheye image: front_000056

DRosemei commented 12 months ago

@yongguanjiangshan You can check the rectify algorithm first. Objects in red boxes are not upright. image

yongguanjiangshan commented 12 months ago

@yongguanjiangshan You can check the rectify algorithm first.

@DRosemei thank you, i have checked the rectify algorithm with my colleague. And modify the code to regenerate the undistort image. finally the result is normal. thanks again!