How to train model of largger scale as input size?

[megvii-basedetection]

How to train model of largger scale as input size?

[yinchimaoliang]

You should change scale_factor in ida accordingly.

[Tony-Hou]

You should change scale_factor in ida accordingly.
ida_aug_conf = { 'resize_lim': (0.386, 0.55), 'final_dim': final_dim, 'rot_lim': (-5.4, 5.4), 'H': H, 'W': W, 'rand_flip': True, 'bot_pct_lim': (0.0, 0.0), 'cams': [ 'CAM_FRONT_LEFT', 'CAM_FRONT', 'CAM_FRONT_RIGHT', 'CAM_BACK_LEFT', 'CAM_BACK', 'CAM_BACK_RIGHT' ], 'Ncams': 6, }
The final_dim map is assigned final_dim variable , Is scale_factor you mean resize_lim ?

[yinchimaoliang]

Yes, I mean resize_lim.

[Tony-Hou]

Yes, I mean resize_lim.
If input size is changed to 896x1600, How to set the resize_lim value corresponding?

[yinchimaoliang]

You may change to [0.95, 1.25].

[Tony-Hou]

You may change to [0.95, 1.25].
Ok, I will try it. What is the reasons that to set [0.95, 1.25] ? It is an experimental parameter?

[wenjie-yuan]

You may change to [0.95, 1.25]
If final_dim=(900,1600), how much should resize_lim be set to? I have tried resize_lim=(1.0,1.0) and (1.0, 1.25), but none of them work. May I ask how the range of resize_lim can be found?