How do you set parameters for finetuning?

Hi authors,

Thank you for the excellent work!

Can I ask for the hyperparameters for finetuning? I want to funetune your model on my own dataset, but no idea how to set the parameters, like, --num_steps, --sample_len, --lr, --train_iters, --valid_iters

python train.py --batch_size 1 \
--spatial_scale -0.2 0.4 --image_size 384 512 --saturation_range 0 1.4 --num_steps 200000 \
--ckpt_path dynamicstereo_sf_dr \
--sample_len 5 --lr 0.0003 --train_iters 10 --valid_iters 20 \
--num_workers 28 --save_freq 100 --update_block_3d --different_update_blocks \
--attention_type self_stereo_temporal_update_time_update_space --train_datasets dynamic_replica things monkaa driving

Start by reading train.py and tracing the listed finetuning arguments, including --num_steps, --sample_len, --lr, --train_iters, and --valid_iters. Compare their defaults and usage with the example command; the work is done when clear guidance for choosing these parameters on a custom dataset is documented.

Written by the indexing model from the issue text.