THTensorCopy copy error about context-encoder

[pathak22]

Could you resolve this error ? Sorry, I noticed this thread today (by chance) and somehow missed the notification earlier.

Few things to try:
=> Print the size of data and datatable in dataset.lua: line 328
=> What if you set this: nc=1, loadSize=129, fineSize=128 ? Does this work ?

[pathak22]

On further thinking, I realized that I have not actually supported grayscale version in code. There are some hard-coded channel-wise things, for example: https://github.com/pathak22/context-encoder/blob/master/train.lua#L283-L285

You will have to edit train.lua etc as well for your code to work on grayscale.

[myw8]

can you tell me where can I get your train data images

[lhaof]

@myw8 https://github.com/tylin/dpatch use this code to download images from google street view

[pathak22]

@myw8 @lhaof Note that currently released code works much better when trained on ImageNet images. For example: see imagenet results on project webpage.

But if you necessarily want to use street-view dataset, you can email me and I will share. Due to license policy issue, it is not allowed to release that dataset publicly.