Yassine Alouini

I guess this issue still needs discussion and there is no point in wanting to contribute a loss for now? :thinking:

Yes, it works for me, thanks @datumbox. 👌 Which one should I pick? Do you have a preference @oke-aditya?

Thanks @oke-aditya. Let me give dIoU a try and see if I can also do the cIoU next. I guess you can help me with review since you know better...

Logistics question: Should I create an issue and add some details or is it enough to start a branch from my work and then do a PR later? @datumbox

@oke-aditya @abhi-glitchhg I think we need to keep our codes synched otherwise we might end up implementing something that is quite different for cIoU and dIoU. I will try to...

Some progress here (it is still a draft): https://github.com/pytorch/vision/pull/5786

Did someone manage to do transfer learning using the DPT large or hybride pretrained models and using another depth dataset? Is the loss described in this [paper](https://papers.nips.cc/paper/2014/file/7bccfde7714a1ebadf06c5f4cea752c1-Paper.pdf) a good one...

@antocad Thanks for this. :ok_hand:

@angrysword I have arrived to the same conclusion trying to train the model on a new dataset. I hope they did open-source the training pipeline but for now will have...

For normalization, I am using an internal scale factor of the camera to move from pixels to meters. Then, I use an estimation of shift and scale factors to align...