Cannot produce results using the pretrained model

[bcv-uniandes]

Cannot produce results using the pretrained model

[changliu19]

By the way, I noticed that the default batch size in the program is set to 1. But when I try to increase the batch size value, I got

RuntimeError: stack expects each tensor to be equal size, but got [3, 512, 512] at entry 0 and [3, 342, 512] at entry 1

It seems that the code kept the aspect ratio of the original image when doing resize so that images in the same batch are of different sizes. I wonder that was your model trained with the batch size of 1? Thank you!

[13331112522]

plus --high-res

[andfoy]

@ntuLC Sorry for the late reply. Please take a look at the discussion of #21 (comment) and see if it solves this one. As @13331112522, you might be missing the --high-res flag, as the pretrained weights are high-resolution ones. Also, please take into account that the dependencies of this project are very old and values may vary with latest releases of PyTorch/SRU.

Regarding the question about the batch size, we had to train our model with a batch size of 1, due to memory constraints and also due to the dynamic convolution computations which do not allow batch dimension computation.