HOU Yuenan
HOU Yuenan
Currently, I have reduced the number of layers from 12 to 10 due to the limited memory and kept other hyperparameters fixed. The searching performance on the Cityscapes validation set...
I have not met that problem. You can just train the model for more epochs and check the result. However, it takes me around 7 days to train the searched...
Please have a look at [this issue](https://github.com/NoamRosenberg/autodeeplab/issues/83).
@Sunshine-Ye Thanks for your sharing. I wonder what is the performance of your searched architecture instead of the provided model? Thanks a lot.
Unfortunately, there is no plan for the update of the codebase. You can try to run the code with newer PyTorch and share your experience with us if there exist...
Example codes on CULane and LLAMAS can be found in [this repo](https://github.com/cardwing/Codes-for-Lane-Detection) and [this repo](https://github.com/cardwing/unsupervised_llamas).
Can you provide the full error log? It seems that libpython3.5m.so.1.0 can not be found by the program and maybe a soft link trick can solve it.
70.2 is the performance of ERFNet-Torch. Since ERFNet-PyTorch is even better than the teacher model, we have to take ERFNet-Torch as student.
Is the low-resolution image from ApolloScape? If yes, you may need to finetune the model. If no, you have to retrain the model from scratch.
Can you try to reimplement the moment pooling part according to the description in our paper?