About shell command of kitti

[wangjiyuan9]

Hi there, this is Amazing work, undoubtedly! But I'd like to ask if you published something wrong in your training command:

$ python main.py --dir_data datta_path --data_name KITTIDC --split_json ../data_json/kitti_dp.json \ --patch_height 352 --patch_width

906

--GPUs 0,1,2,3 --loss 1.0L1+1.0L2+1.0*DDIM --epochs 30 \ --batch_size 8 --max_depth 88.0 --save NAME_TO_SAVE \ --model_name DiffusionDCbase --backbone_module swin --backbone_name swin_large_naive_l4w722422k --head_specify DDIMDepthEstimate_Swin_ADDHAHI

In your paper there are only 706*352, what is 906 ?

[duanyiqun]

Hi thanks for interest.

This is actually crop_widthcrop_height of input image. The final version used 706352 size for training. However, for train from scratch 906 is more stable.

[wangjiyuan9]

Thanks for you reply!