search

isl-org / ZoeDepth

Metric depth estimation from a single image
MIT License
2.25k stars 208 forks source link

Please update documentation with more details on training the other midas backbones #52

Open MalekWahidi opened 1 year ago

MalekWahidi commented 1 year ago

I have been trying to train a new ZoeDepth_N model on the NYUv2 dataset with the more efficient DPT_SwinV2_L_384 MiDaS backbone for real-time performance. However, it is not clear from the current documentation about how to properly setup the dataset and config parameters. I have no idea where to find the "shortcuts/datasets/nyu_depth_v2/official_splits/test/". The train_mono script is just printing the config params and does nothing else. What am I missing?

Shiyao-Xu commented 1 year ago

Have the same question here.

Gilgamesh666666 commented 1 year ago

me too

Gilgamesh666666 commented 1 year ago

I find the solution here https://github.com/cleinc/bts/tree/master#prepare-nyu-depth-v2-test-set. you can follow the instruction to download the NYU mat file and then copy the python file into your own directory and download the split.mat in this github.

MalekWahidi commented 1 year ago

Yes thanks, this worked for me too.

Shiyao-Xu commented 1 year ago

I find the solution here https://github.com/cleinc/bts/tree/master#prepare-nyu-depth-v2-test-set. you can follow the instruction to download the NYU mat file and then copy the python file into your own directory and download the split.mat in this github.

Thanks for sharing! Yet I still got error: only integer scalar arrays can be converted to a scalar index. Seems from h5py. Any idea how to fix that?

AlexMousaei commented 11 months ago

@MalekWahidi I'm new to this what is required to switch the backbone? Given that you've navigated already, I was wondering if you could share some insights or a brief step-by-step guide on how you made the switch? Thank you.

AlexMousaei commented 11 months ago

I find the solution here https://github.com/cleinc/bts/tree/master#prepare-nyu-depth-v2-test-set. you can follow the instruction to download the NYU mat file and then copy the python file into your own directory and download the split.mat in this github.

Thanks for sharing! Yet I still got error: only integer scalar arrays can be converted to a scalar index. Seems from h5py. Any idea how to fix that?

@Shiyao-Xu In "extract_official_train_test_set_from_mat.py" Replace line 87: scenes = [u''.join(chr(c) for c in h5_file[obj_ref]) for obj_ref in h5_file['sceneTypes'][0]] With: scenes = [u''.join(chr(c[0]) for c in h5_file[obj_ref][:]) for obj_ref in h5_file['sceneTypes'][0]]

baibizhe commented 10 months ago

scenes = [u''.join(chr(c[0]) for c in h5_file[obj_ref][:]) for obj_ref in h5_file['sceneTypes'][0]]

Thanks!