Open Serenade-ajp opened 2 years ago
More strangely, another 640x1600 model behaves in the opposite way And the only difference of the two models is : the first one use bev res 256x256, while the later one use 128x128
Have you changed train_cfg
, test_cfg
and bbox_coder
accordingly?
You mean changing 'out_size_factor' when change bev res from 128x128 to 256x256 ? we have done this. what is strange is that some cam fov is normal while others are not good, if some config params are wrongly set, what kind of params can make the different behavior of different cams?
And the 640x1600, 128x128 model is modified from your 'bev_depth_lss_r50_256x704_128x128_20e_cbgs_2key.py', the only difference is change following params: final_dim = (640, 1600) backbone_conf['final_dim'] = final_dim ida_aug_conf['final_dim'] = final_dim ida_aug_conf['resize_lim'] = (0.94, 1.25)
The config params of 640x1600 model is as follows: final_dim = (640, 1600) backbone_conf['final_dim'] = final_dim ida_aug_conf['final_dim'] = final_dim ida_aug_conf['resize_lim'] = (0.94, 1.25) The inference results is strange, only the instance in CAM_BACK fov seems to be normal, while objects in other cam's fov tend to miss a certain scale
Hi~ your visualization is awesome, but I didn't find any visualization in BEVDepth's repo. Is your visualization codes available? thx a lot
Hi! I want to know where (which file?) you change the bev res from 128x128 to 256x256 ? I can not found it T T. Best wishes!!
And the 640x1600, 128x128 model is modified from your 'bev_depth_lss_r50_256x704_128x128_20e_cbgs_2key.py', the only difference is change following params: final_dim = (640, 1600) backbone_conf['final_dim'] = final_dim ida_aug_conf['final_dim'] = final_dim ida_aug_conf['resize_lim'] = (0.94, 1.25)
Hello, I also encountered this problem and only made the same modification on the original code. Have you solved this problem? I have been troubled by this problem for a long time, looking forward to your reply.
The config params of 640x1600 model is as follows: final_dim = (640, 1600) backbone_conf['final_dim'] = final_dim ida_aug_conf['final_dim'] = final_dim ida_aug_conf['resize_lim'] = (0.94, 1.25) The inference results is strange, only the instance in CAM_BACK fov seems to be normal, while objects in other cam's fov tend to miss a certain scale