noahzn / Lite-Mono

[CVPR2023] Lite-Mono: A Lightweight CNN and Transformer Architecture for Self-Supervised Monocular Depth Estimation
MIT License
540 stars 61 forks source link

On the depth map and true distance obtained from the model #80

Closed yinyang472 closed 1 year ago

yinyang472 commented 1 year ago

I would like to know the conversion relationship between the depth map obtained from your model and the actual distance. Can you please let me know

noahzn commented 1 year ago

Hi, all the predicted depth values are relative depth. We can not get the actual distance using this monocular method.

yinyang472 commented 1 year ago

but the dataset you used are actual distance

noahzn commented 1 year ago

but the dataset you used are actual distance

Hi, no. We didn't use the ground-truth during training. During evaluation, we use median scaling to scale the depth.

yinyang472 commented 1 year ago

thank you for your answers. by the way I am trying to convert the the outcome your model give to ground-truth. should I find a References

noahzn commented 1 year ago

Yes, you can use median scaling to find a scale ratio and multiply each value by that ratio. This is implemented in the evaluation script, see this.

yinyang472 commented 1 year ago

thank you 🙂