search

isl-org / DPT

Dense Prediction Transformers
MIT License
1.96k stars 254 forks source link

Training and evaluation code #64

Open harsh-sengar opened 2 years ago

harsh-sengar commented 2 years ago

You haven't released the training and evaluation code for Monodepth and Semantic Segmentation tasks. If you can provide your training and evaluation code then that will be very useful for my research.

younesbelkada commented 2 years ago

Hi, we have made an attempt to reproduce the training script of DPT for semantic segmentation and depth estimation in the following repo: https://github.com/antocad/FocusOnDepth check it out if you are interested!

awsaf49 commented 2 years ago

@younesbelkada Did you try finetuning it on kitti dataset as well?

younesbelkada commented 2 years ago

Hi @awsaf49 unfortunately we didn't, but I think that it should work well if you follow the instructions for training

awsaf49 commented 2 years ago

@younesbelkada Thanks for your reply. The issue with kitti is that it doesn't come with dense/interpolated depth map like nyuv2 dataset. Hence, I'm having doubts if they trained model with sparse depths or dense/interpolated depth

jikexiaohgege commented 3 months ago

你好，我们尝试在以下 repo 中重现用于语义分割和深度估计的 DPT 训练脚本：https://github.com/antocad/FocusOnDepth，如果您有兴趣，请查看！

您好 请问您提供的这个代码是如何在作者这个模型中训练的，感谢指导