lhoyer
about'multi-task learning'
hello， i'm very interested in the paper. I have some questions about the execution of the entire process. 1."python train.py --machine ws --config configs/cityscapes_monodepth_highres_dec5_crop.yml python train.py --machine ws --config configs/cityscapes_monodepth_highres_dec6_crop.yml" Are these two commands referring to the part that says, "During the first 300k iterations, only the depth decoder and the pose network are trained. Afterwards, the depth encoder is fine-tuned with an ImageNet feature distance λF = 1 × 10−2 for another 50k iterations"? 2.What is the difference between "run_experiments.py" and "train.py"? If I want to follow the training process described in the "Experiments.Training" section of the paper to train the depth network, pose network, and semantic segmentation network, how should I proceed?
hello,
I noticed that in train.py, even though there's a function defined as "def save_monodepth_models(self)", this function is not used within the training process.
hello,
What I want to know is, if the 'save_monodepth_models' function is not used, how can we obtain the .pth file during depth pretraining?