differences between fov-mono3d and mv-mono3d

[hanssssssss]

Hi, I am confused about the difference between the fov-mono3d and mv-mono3d on the waymo dataset. I found that the number of data being trained is less than that in the dataset. Could you please me more information about that? Thanks.

[ZCMax]

fov-mono3d means we only use the front view camera images for training mono3d model, mv-mono3d means we use all the camera images for training mono3d model

[hanssssssss]

Thanks for the quick reply. I am also confused about the number of data being trained. For instance, the number of data per epoch under mv-mono3d setting on waymo dataset is 7905*batchsize(3), which is less than the number of data in any folder(image_0, image_1 ....) in the folder.
besides, in my understanding, image_0 folder should be the front view image, but the the number of data being trained can not match the number of data in the folder.

[ZCMax]

the load_interval is set to be 5, meaning using one image every 5 images. Besides, the batch_size is not 3 during training, if you are using the config of https://github.com/open-mmlab/mmdetection3d/blob/dev-1.x/configs/pgd/pgd_r101_fpn-head_dcn_16xb3_waymoD5-mv-mono3d.py, the real batch_size should be 16x3=48.

[hanssssssss]

Thanks! Btw, I did not use the default config files.