Training with Cityscapes datasets

xy-guo / Learning-Monocular-Depth-by-Stereo

Learning Monocular Depth by Distilling Cross-domain Stereo Networks, ECCV18

https://arxiv.org/abs/1808.06586

MIT License

94 stars 8 forks source link

Training with Cityscapes datasets #10

Closed sazippe closed 5 years ago

sazippe commented 5 years ago

Table 1 StereoUnsupFt→Mono pt No S,K→K 0.099 0.745 4.424 0.182 0.884 0.963 0.983 StereoUnsupFt→Mono pt No S,K→C,K 0.095 0.703 4.316 0.177 0.892 0.966 0.984

From the result in Table1, i can see that using Cityscapes datasets will improve overall performance.

I want to know how to use Cityscapes datasets on training. After training mono network by KITTI datasets, then finetune its network by Cityscapes datasets? Or training mono network by both KITTI and Cityscapes datasets?

xy-guo commented 5 years ago

You can use the StereoNoFt model to pretrain the monocular model on Cityscapes. Remember to cut off the bottom part for the Cityscapes dataset.

sazippe commented 5 years ago

I want to know the flow of the following training. StereoUnsupFt→Mono pt No S,K→C,K

Is this flow correct? First, train stereo model on scenflow and then finetune with KITTI. Next, train mono model (pt with VGG) on Cityspace and then finetune with KITTI.

xy-guo commented 5 years ago

Yes

sazippe commented 5 years ago

Thank you! Training list is 'list/cityscapes_train_list.txt', right?

xy-guo commented 5 years ago

Yes. Note that you should supervise the mono model (pt with VGG) on Cityscapes with StereoNoFt and then finetune on KITTI with StereoUnsup/SupFt model. The learning rate is not changed for finetuning.

sazippe commented 5 years ago

Thank you for your advice.