Train Image Captioning Code on Other Datasets about pytorch-tutorial HOT 7 CLOSED

@ahkarami No, it does not support. However, you can solve the problem by using the DataParallel module, or you can spread modules over multiple GPUs by passing an additional argument to .cuda(). For example, .cuda(1) will place the tensors and modules on 2nd GPU. For this, you should edit here and here.

from pytorch-tutorial.

@ahkarami Sorry for late reply. You can prepare a JSON file in the same form as MSCOCO dataset and use the code in the tutorial. Another way is to implement your own DataSet class. See here for how to implement custom dataset in PyTorch.

from pytorch-tutorial.

Dear @yunjey,
Thank you very much for your response. Just as another question, would you please tell me that your nice Image Captioning code supports Multi-GPU training or not?

from pytorch-tutorial.

Dear @yunjey, Thank you very much for your time and useful response.

from pytorch-tutorial.

@ahkarami @yunjey hi guys i am new to this topic i need to train for a custom dataset can you pls elaborate the steps for training with custom datasets . These are the things i am having currently with me

1.10k img database of various objects and tags for each image

Thanks in advance

from pytorch-tutorial.

Dear @abhigoku10,
You can use image captioning codebase in pytorch instead of above code. Actually, as I remember, I haven't use above code for training on another data sets, but by using the mentioned code I have done that. However, as I have remembered, that was a semi-sophisticated work to create a json file like the used one for MS-COCO.

from pytorch-tutorial.

Dear @yunjey
I want to train the model on my dataset, but it seems that my json file works wrong, could you tell me how to preprocess a json file?
Thanks very much

from pytorch-tutorial.