search

researchmm / PEN-Net-for-Inpainting

[CVPR'2019] PEN-Net: Learning Pyramid-Context Encoder Network for High-Quality Image Inpainting
https://arxiv.org/abs/1904.07475
MIT License
361 stars 77 forks source link

about ATNConv #11

Open XiaoYangon opened 4 years ago

XiaoYangon commented 4 years ago

Hi , I have a question. in the ATNConv module, the patches of high level feature map are got with stride 1，Is the calculation too large in conv，your gpu config？，Can it be equal to 2 or larger number？

zengyh1900 commented 4 years ago

HI, since the spatial size of the highest level of feature map is relatively small, so we use stride 1, which is more suitable. I think you can try a larger stride. But I think the results would not be better due to the small spatial size of the high-level feature maps.

Thanks, Yanhong.

XiaoYangon commented 4 years ago

hello，I want to use places2 data to trian the model, each class is a zipfile from a to z in places2 data. In training process, the data is from zipfile according your dataset.py. I use two 1080Ti GPUs, but the error is appear： spawn.py 167 spawn while not spawn_context.join(): spawn.py 103 join (error_index, name) Exception:process 1 terminated with signal SIGKILL I don't know why, can you tell me about your gpu config?