search

DepthAnything / Depth-Anything-V2

Depth Anything V2. A More Capable Foundation Model for Monocular Depth Estimation
https://depth-anything-v2.github.io
Apache License 2.0
3.39k stars 277 forks source link

use bash dist_train.sh to train on my own datasets, the results are bad #86

Open visonpon opened 2 months ago

visonpon commented 2 months ago

hi @LiheYoung , I tried to use dist_train.sh to train on my own datasets, but the results get worse, I wonder if there are some preprocess i ignored or exist some tricks for fine-tune.

Hope you can give some advice, thanks~

Edric-star commented 2 months ago

Hi there, sorry to bother you but would you mind sharing me with some of your details while fine-tuning on your own datasets? Because I was doing it as well and confronted with some problems. @visonpon

visonpon commented 2 months ago

you just need to follow thedataets/kitti.py script. btw, my training starts normally. but as training goes, predicted results get worse.

Edric-star commented 2 months ago

you just need to follow thedataets/kitti.py script. btw, my training starts normally. but as training goes, predicted results get worse.

Hi there, my results are not good and thanks I followed kitti datasets since my gt are sparse as well

shilpaullas97 commented 1 week ago

you just need to follow thedataets/kitti.py script. btw, my training starts normally. but as training goes, predicted results get worse.

Hi @visonpon ,

I'm facing similar issue with metric depth training on custom dataset. Could you please let me know the steps followed in details? Were you able to resolve it then?