Shariq F. Bhat

Hi Divadi, Thanks for your interest in our work. This is an interesting question. Since the self-attention layers combine information globally (from all tokens to all tokens), there's actually no...

Hi, Thanks for your interest in our work. I hypothesize that changing `min_val` and `max_val` on pretrained model would just change units the output depth map uses. By default, the...

This is because there are only 500 learned positional encodings and if you try to infer an image much higher than the default model resolution, then the number of tokens...

Please refer to instructions provided [here](https://github.com/facebookresearch/pytorch3d/blob/main/INSTALL.md) to install pytorch3d. If you can't install pytorch3d for your driver version, you may also give a try to pytorch3d-nightly. As @eugenelyj pointed out,...

DenseDepth uses inpainted ground truth depth maps (required particularly for SSIM loss) which may result in noisy supervision, while as AdaBins is trained on raw ground truth depth maps (which...

Hi, Thank you for pointing this out. This might actually be a bug (also present when AdaBins was trained). OnecycleLR Scheduler should get the lrs for all param groups as...

It's probably an issue with wandb in a distributed setting. Try calling `wandb.finish()` at the end of main worker (i.e. in every process) to let the wandb know its time...

> Dear authors, > > > > I'm trying to reproduce your results on NYU Depth V2 dataset, but I'm facing some problems regarding the evaluation results, both retraining the...

@leoshine Different from ours, DenseDepth uses inpainted depth maps. Please refer to [this](https://github.com/shariqfarooq123/AdaBins/issues/54#issuecomment-1014929303) for details.